Hi,
Ok I was looking at this learning challenge,
producing vector (y1,y2,y3,y4) from a vector
(x1,x2,x3,x4), System R can do it via least square?
| 0 0 0 1 |-a-a | x1 |-a-a-a-a | x4 |
| 0 0 1 0 |-a-a | x2 |-a =-a | x3 |
| 0 1 0 0 |-a-a | x3 |-a-a-a-a | x2 |
| 1 0 0 0 |-a-a | x4 |-a-a-a-a | x1 |
How it started:
"multiplicative RNNs arises naturally from a
proof-theoretic interpretation of next-token
prediction as nested intuitionistic implication"
Paul Tarau - 2026
https://arxiv.org/abs/2601.19915
How its going:
"Dave uses a PDP-11 to train a real Neural
Network complete with Transformers and
Attention so you can see them at their most basic."
Mr. Taskmanager - 2026
https://www.youtube.com/watch?v=OUE3FSIk46g
We see Doctor Frankstein in action from
the Bronze Age of Computing, producing
a Humunkulus, the progenitor of todays
Bulgakov Shuriks in the Hyperscale Age!
Bye
P.S.: My impression neither cut to the core, that
this incredible transformer most likely
produced this deterministic attention:
| -1 | * | k | + | 5 | = | k' |
Or differently expressed y_k = x_{5-k}.
How did the transformer do it? It produced
a neural network with 1216 parameters, but
didn't use embeddings or polar encoding
of positions. But if we strip the noise
and denoise from the position encoding,
the denoise is done via softmax. We somehow
must get the above, right? I still need to
verify my claim! BTW: The PDP-11 assembly
from 1979 uses wider example not with n=4
but with n=8.
On 7/19/2026 2:04 PM, Chris M. Thomasson wrote:
I am friends with Dmitry Vyukov from way back, 23+ years ago. Actually, I helped him find some bugs in Relacy when it was in pre-alpha way back on comp.programming.threads. Also, I created some neat eventcount algos, read all:
https://gist.github.com/mratsim/04a29bdd98d6295acda4d0677c4d0041
For starters... ;^)
also, I made a neat alteration to one of Dmitry Vyukov MPMC queues, check it out:
https://groups.google.com/g/lock-free/c/acjQ3-89abE/m/a6-Di0GZsyEJ
can you get to the link? Thanks. Read all.
Hi,
I'm a spinner, I'm a sinner
I spin on CAS loops for my dinner
Some call it busy-wait, I call it fate
When the queue is empty, I just rotate
2 Producer threads, 2 Consumer threads
Each Producer generating 8192 items
the single common channel between
them 128 elements large.
public class DmitryVyukov
8 ms
public class DougLea
10 ms
On the Ryzen AI 7 350 which has 8 physical
cores with hyperthreading, making it 16
logical cores.
LoL
Bye
See also:
Producer-Consumer Queues by Dmitry Vyukov https://sites.google.com/site/1024cores/home/lock-free-algorithms/queues
Steve Miller Band - The Joker (Official Music Video) https://www.youtube.com/watch?v=dV3AziKTBUo
Mild Shock schrieb:
Hi,
Ok I was looking at this learning challenge,
producing vector (y1,y2,y3,y4) from a vector
(x1,x2,x3,x4), System R can do it via least square?
| 0 0 0 1 |-a-a | x1 |-a-a-a-a | x4 |
| 0 0 1 0 |-a-a | x2 |-a =-a | x3 |
| 0 1 0 0 |-a-a | x3 |-a-a-a-a | x2 |
| 1 0 0 0 |-a-a | x4 |-a-a-a-a | x1 |
How it started:
"multiplicative RNNs arises naturally from a
proof-theoretic interpretation of next-token
prediction as nested intuitionistic implication"
Paul Tarau - 2026
https://arxiv.org/abs/2601.19915
How its going:
"Dave uses a PDP-11 to train a real Neural
Network complete with Transformers and
Attention so you can see them at their most basic."
Mr. Taskmanager - 2026
https://www.youtube.com/watch?v=OUE3FSIk46g
We see Doctor Frankstein in action from
the Bronze Age of Computing, producing
a Humunkulus, the progenitor of todays
Bulgakov Shuriks in the Hyperscale Age!
Bye
P.S.: My impression neither cut to the core, that
this incredible transformer most likely
produced this deterministic attention:
| -1 | * | k | + | 5 | = | k' |
Or differently expressed y_k = x_{5-k}.
How did the transformer do it? It produced
a neural network with 1216 parameters, but
didn't use embeddings or polar encoding
of positions. But if we strip the noise
and denoise from the position encoding,
the denoise is done via softmax. We somehow
must get the above, right? I still need to
verify my claim! BTW: The PDP-11 assembly
from 1979 uses wider example not with n=4
but with n=8.
completely nonsense. Not even correct grammatically
Ross Finlayson wrote:
access to memory, has that these days with
L1/L2/L3 caches and the proximity and affinity
in the topology of the cores and
those are embedded cpu registers area, it has
nothing to do with the ram. Works by higher
clocks compared, but merely useless in AI.
Mild Shock wrote:
me about lazySet() in Java. But I went with set():
private static boolean enqueue(Queue q, Object data) {
int pos = q.enqueuePos.get();
for (; ; ) {
int index = pos & q.bufferMask; int seq =
q.sequences.get(index);
int dif = seq - pos;
if (dif == 0) {
if (q.enqueuePos.compareAndSet(pos, pos + 1)) {
q.data[index] = data; q.sequences.set(index, pos +
1);
return true;
}
pos = q.enqueuePos.get();
} else if (dif < 0) {
return false;
} else {
pos = q.enqueuePos.get();
}
}
}
The above version seems to be more suitable for my purpose, since it
allows polling, it basically implements offer(). While the
completely nonsense. Not even correct grammatically
Hi,
completely nonsense. Not even correct grammatically
Yes of course, Micro Penis will know. Nothing, LoL
But here we find how village idiot Micro Penis is.
The Cache Identity Crisis by Micro Penis:
Claiming that L1/L2/L3 caches and core
topology are "embedded cpu registers area,
it has nothing to do with the ram." That is
a brilliant piece of hardware fiction.
Caches are SRAM acting as a hierarchical
staging ground for and backed by system
or device RAM, not an alternate dimension
divorced from memory entirely.
And calling cache hierarchies "registers"
is like confusing a warehouse parking lot
with the glove compartment of a single car.
Never heard of the crossbar in AMD GPUs?
LoL
Bye
Obelin Baisaroff schrieb:
Ross Finlayson wrote:
access to memory, has that these days with
L1/L2/L3 caches and the proximity and affinity
in the topology of the cores and
those are embedded cpu registers area, it has
nothing to do with the ram. Works by higher
clocks compared, but merely useless in AI.
Romelio Balakhonsky schrieb:
Mild Shock wrote:
pos +me about lazySet() in Java. But I went with set():
-a-a-a-a-a-a private static boolean enqueue(Queue q, Object data) {
-a-a-a-a-a-a-a-a-a-a int pos = q.enqueuePos.get();
-a-a-a-a-a-a-a-a-a-a for (; ; ) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a int index = pos & q.bufferMask; int seq =
-a-a-a-a-a-a-a-a-a-a-a-a-a-a q.sequences.get(index);
-a-a-a-a-a-a-a-a-a-a-a-a-a-a int dif = seq - pos;
-a-a-a-a-a-a-a-a-a-a-a-a-a-a if (dif == 0) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a if (q.enqueuePos.compareAndSet(pos, pos + 1)) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a q.data[index] = data; q.sequences.set(index,
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a 1);
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a return true;
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a }
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a pos = q.enqueuePos.get();
-a-a-a-a-a-a-a-a-a-a-a-a-a-a } else if (dif < 0) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a return false;
-a-a-a-a-a-a-a-a-a-a-a-a-a-a } else {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a pos = q.enqueuePos.get();
-a-a-a-a-a-a-a-a-a-a-a-a-a-a }
-a-a-a-a-a-a-a-a-a-a }
-a-a-a-a-a-a }
The above version seems to be more suitable for my purpose, since it
allows polling, it basically implements offer(). While the
completely nonsense. Not even correct grammatically
Hi,
If you don't know how GPU caches work.
Just RTFM the RDNA 3.5 specs! They have
some explanations in the docs, what jobs
the GPU cache lines do, in relation to
what instructions:
Abbreviation for rCyRead The Fucking ManualrCO. http://www.catb.org/esr/jargon/html/R/RTFM.html
Here some RDNA 4.0 specs (smaller GPUs):
"RDNA4" Instruction Set Architecture
Reference Guide - 7-April-2025 https://docs.amd.com/v/u/en-US/rdna4-instruction-set-architecture
Here some CDNA 4.0 specs (bigger GPUs):
CDNA4 Instruction Set Architecture
Reference Guide - 5-August-2025 https://www.amd.com/content/dam/amd/en/documents/instinct-tech-docs/instruction-set-architectures/amd-instinct-cdna4-instruction-set-architecture.pdf
Bye
But the difference between CDNA GPUs and
RDNA GPUs seems not to be that big, at least
in 2025:
Commitment to High-Performance
Computing in the Age of AI https://docs.amd.com/v/u/en-US/precision-at-scale-amd-whitepaper
While the Copilot+ threshold is 45 TFLOPs,
a MI355X, Datacenter, rack-mounted, liquid cooling,
has surely more memory, but only 157 TFLOPS.
And RTX 5090, Desktop, 600W+ TDP, massive cooling,
has also only 100 TFLOPS. So it looks these
Copilot+ AI Laptops are pretty swell, arent they?
Mild Shock schrieb:
Hi,
completely nonsense. Not even correct grammatically
Yes of course, Micro Penis will know. Nothing, LoL
But here we find how village idiot Micro Penis is.
The Cache Identity Crisis by Micro Penis:
Claiming that L1/L2/L3 caches and core
topology are "embedded cpu registers area,
it has nothing to do with the ram." That is
a brilliant piece of hardware fiction.
Caches are SRAM acting as a hierarchical
staging ground for and backed by system
or device RAM, not an alternate dimension
divorced from memory entirely.
And calling cache hierarchies "registers"
is like confusing a warehouse parking lot
with the glove compartment of a single car.
Never heard of the crossbar in AMD GPUs?
LoL
Bye
Obelin Baisaroff schrieb:
Ross Finlayson wrote:
access to memory, has that these days with
L1/L2/L3 caches and the proximity and affinity
in the topology of the cores and
those are embedded cpu registers area, it has
nothing to do with the ram. Works by higher
clocks compared, but merely useless in AI.
Romelio Balakhonsky schrieb:
Mild Shock wrote:pos +
me about lazySet() in Java. But I went with set():
-a-a-a-a-a-a private static boolean enqueue(Queue q, Object data) {
-a-a-a-a-a-a-a-a-a-a int pos = q.enqueuePos.get();
-a-a-a-a-a-a-a-a-a-a for (; ; ) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a int index = pos & q.bufferMask; int seq = >> -a>>-a-a-a-a-a-a-a-a-a-a-a-a-a-a q.sequences.get(index);
-a-a-a-a-a-a-a-a-a-a-a-a-a-a int dif = seq - pos;
-a-a-a-a-a-a-a-a-a-a-a-a-a-a if (dif == 0) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a if (q.enqueuePos.compareAndSet(pos, pos + 1)) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a q.data[index] = data; q.sequences.set(index,
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a 1);
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a return true;
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a }
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a pos = q.enqueuePos.get();
-a-a-a-a-a-a-a-a-a-a-a-a-a-a } else if (dif < 0) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a return false;
-a-a-a-a-a-a-a-a-a-a-a-a-a-a } else {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a pos = q.enqueuePos.get();
-a-a-a-a-a-a-a-a-a-a-a-a-a-a }
-a-a-a-a-a-a-a-a-a-a }
-a-a-a-a-a-a }
The above version seems to be more suitable for my purpose, since it
allows polling, it basically implements offer(). While the
completely nonsense. Not even correct grammatically
Mild Shock wrote:immediately
Yes of course, Micro Penis will know. Nothing, LoL But here we find how
village idiot Micro Penis is.
The Cache Identity Crisis by Micro Penis:
Claiming that L1/L2/L3 caches and core topology are "embedded cpu
registers area,
it has nothing to do with the ram." That is a brilliant piece of
hardware fiction.
Caches are SRAM acting as a hierarchical
imbecile, formatting code like that puts you in prison in normal
countries. You are using entire lines for single chars, ie "}" etc, imbecile.
also, fucking lying illiterate imbecile
L1, L2, and L3 caches are located on the CPU, not in the system RAM. They form a high-speed memory hierarchy embedded directly within or
adjacent to the processor cores to minimize data access latency.
Hi,
You are a moron right? Modern GPU has its
own L1,..,Ln caches. RTFM ASAP you brain
amputated vodka alcoholic.
Russian Hardbass - Crazy Dance
https://www.youtube.com/watch?v=6OoglRuT0jY
Bye
Jeiker Makulov schrieb:
Mild Shock wrote:
Yes of course, Micro Penis will know. Nothing, LoL But here we find how
village idiot Micro Penis is.
The Cache Identity Crisis by Micro Penis:
Claiming that L1/L2/L3 caches and core topology are "embedded cpu
registers area,
it has nothing to do with the ram." That is a brilliant piece of
hardware fiction.
Caches are SRAM acting as a hierarchical
imbecile, formatting code like that puts you in prison in normal countries. You are using entire lines for single chars, ie-a "}" etc, imbecile.
also, fucking lying illiterate imbecile
L1, L2, and L3 caches are located on the CPU, not in the system RAM.They
form a high-speed memory hierarchy embedded directly within orimmediately
adjacent to the processor cores to minimize data access latency.
Hi,
If you don't know how GPU caches work.
Just RTFM the RDNA 3.5 specs! They have
some explanations in the docs, what jobs
the GPU cache lines do, in relation to
what instructions:
Abbreviation for rCyRead The Fucking ManualrCO. http://www.catb.org/esr/jargon/html/R/RTFM.html
Here some RDNA 4.0 specs (smaller GPUs):
"RDNA4" Instruction Set Architecture
Reference Guide - 7-April-2025 https://docs.amd.com/v/u/en-US/rdna4-instruction-set-architecture
Here some CDNA 4.0 specs (bigger GPUs):
CDNA4 Instruction Set Architecture
Reference Guide - 5-August-2025 https://www.amd.com/content/dam/amd/en/documents/instinct-tech-docs/instruction-set-architectures/amd-instinct-cdna4-instruction-set-architecture.pdf
Bye
But the difference between CDNA GPUs and
RDNA GPUs seems not to be that big, at least
in 2025:
Commitment to High-Performance
Computing in the Age of AI https://docs.amd.com/v/u/en-US/precision-at-scale-amd-whitepaper
While the Copilot+ threshold is 45 TFLOPs,
a MI355X, Datacenter, rack-mounted, liquid cooling,
has surely more memory, but only 157 TFLOPS.
And RTX 5090, Desktop, 600W+ TDP, massive cooling,
has also only 100 TFLOPS. So it looks these
Copilot+ AI Laptops are pretty swell, arent they?
Mild Shock schrieb:
Hi,
completely nonsense. Not even correct grammatically
Yes of course, Micro Penis will know. Nothing, LoL
But here we find how village idiot Micro Penis is.
The Cache Identity Crisis by Micro Penis:
Claiming that L1/L2/L3 caches and core
topology are "embedded cpu registers area,
it has nothing to do with the ram." That is
a brilliant piece of hardware fiction.
Caches are SRAM acting as a hierarchical
staging ground for and backed by system
or device RAM, not an alternate dimension
divorced from memory entirely.
And calling cache hierarchies "registers"
is like confusing a warehouse parking lot
with the glove compartment of a single car.
Never heard of the crossbar in AMD GPUs?
LoL
Bye
Obelin Baisaroff schrieb:
Ross Finlayson wrote:
access to memory, has that these days with
L1/L2/L3 caches and the proximity and affinity
in the topology of the cores and
those are embedded cpu registers area, it has
nothing to do with the ram. Works by higher
clocks compared, but merely useless in AI.
Romelio Balakhonsky schrieb:
Mild Shock wrote:pos +
me about lazySet() in Java. But I went with set():
private static boolean enqueue(Queue q, Object data) {
int pos = q.enqueuePos.get();
for (; ; ) {
int index = pos & q.bufferMask; int seq =
q.sequences.get(index);
int dif = seq - pos;
if (dif == 0) {
if (q.enqueuePos.compareAndSet(pos, pos + 1)) {
q.data[index] = data; q.sequences.set(index,
1);
return true;
}
pos = q.enqueuePos.get();
} else if (dif < 0) {
return false;
} else {
pos = q.enqueuePos.get();
}
}
}
The above version seems to be more suitable for my purpose, since it
allows polling, it basically implements offer(). While the
completely nonsense. Not even correct grammatically
That's too located close to the same cpu, idiot.
Hi,
Just read something for once you alcoholic:
GPU Cache Hierarchy: Understanding L1, L2, and VRAM https://charlesgrassi.dev/blog/gpu-cache-hierarchy/
Measuring Cache Efficiency
GPU profilers like NVIDIA Nsight, AMD Radeon
GPU Profiler, and RenderDoc expose cache hit
rates and memory throughput. Key metrics to watch:
- L1 Hit Rate: Should be >80% for well-optimized
-a-a-a shaders. Below 60% indicates access pattern problems.
- L2 Hit Rate: Measures working set fit.
-a-a-a Below 70% suggests texture thrashing or too many unique textures.
- Texture Memory Throughput: Compare to theoretical
-a-a-a max. Low throughput with high latency = cache misses.
- Memory-bound vs Compute-bound: If memory is the
-a-a-a bottleneck, cache optimization has highest impact.
If the GPU is a discrete GPU, it has VRAM,
If the GPU is an accelerator processing unit (APU),
it shares the system memory with the CPU.
In the APU scenrio often advanced unified memory
is used, which can be multi channel for example.
The high L1 and L2 hit rate for shaders, translates
also to pi-WAM. Which shows, in the example I used,
a quite good GPU saturation, less memory bound and
more compute bound. I have compute these figures already.
Bye
Mild Shock schrieb:
Hi,
You are a moron right? Modern GPU has its
own L1,..,Ln caches. RTFM ASAP you brain
amputated vodka alcoholic.
Russian Hardbass - Crazy Dance
https://www.youtube.com/watch?v=6OoglRuT0jY
Bye
Jeiker Makulov schrieb:
Mild Shock wrote:find how
Yes of course, Micro Penis will know. Nothing, LoL But here we
RAM. Theyvillage idiot Micro Penis is.
The Cache Identity Crisis by Micro Penis:
Claiming that L1/L2/L3 caches and core topology are "embedded cpu
registers area,
it has nothing to do with the ram." That is a brilliant piece of
hardware fiction.
Caches are SRAM acting as a hierarchical
imbecile, formatting code like that puts you in prison in normal
countries. You are using entire lines for single chars, ie-a "}" etc,
imbecile.
also, fucking lying illiterate imbecile
L1, L2, and L3 caches are located on the CPU, not in the system
form a high-speed memory hierarchy embedded directly within orimmediately
adjacent to the processor cores to minimize data access latency.
Hi,
That's too located close to the same cpu, idiot.
Yes, colocating a CPU and GPU onto the same
processor die or package provides massive b
enefits, primarily by eliminating the latency
and bandwidth bottlenecks of moving data
across a motherboard.
Individual L1 Caches: Yes, the CPU cores and
the GPU compute units (Streaming Multiprocessors /
Compute Units) still maintain private,
individual L1 caches.
The "Colocation" Benefit: In traditional
setups (a separate CPU and GPU on PCIe slots),
data must be copied from system RAM to the
GPU's VRAM. When a CPU and GPU are colocated
(like in Apple Silicon, AMD APUs, or Intel
processors), they share a Unified Memory
Architecture (UMA). Both processors can read
and write to the exact same physical memory
pool without redundant data copying
Shared Last-Level Cache: While the L1 caches
remain separate, colocated architectures
often share a high-capacity Last-Level
Cache (L2 or L3). This allows the CPU and
GPU to quickly pass data back and forth
locally on the chip.
Bye
Mild Shock schrieb:
Hi,
Just read something for once you alcoholic:
GPU Cache Hierarchy: Understanding L1, L2, and VRAM
https://charlesgrassi.dev/blog/gpu-cache-hierarchy/
Measuring Cache Efficiency
GPU profilers like NVIDIA Nsight, AMD Radeon
GPU Profiler, and RenderDoc expose cache hit
rates and memory throughput. Key metrics to watch:
- L1 Hit Rate: Should be >80% for well-optimized
-a-a-a-a shaders. Below 60% indicates access pattern problems.
- L2 Hit Rate: Measures working set fit.
-a-a-a-a Below 70% suggests texture thrashing or too many unique textures. >> - Texture Memory Throughput: Compare to theoretical
-a-a-a-a max. Low throughput with high latency = cache misses.
- Memory-bound vs Compute-bound: If memory is the
-a-a-a-a bottleneck, cache optimization has highest impact.
If the GPU is a discrete GPU, it has VRAM,
If the GPU is an accelerator processing unit (APU),
it shares the system memory with the CPU.
In the APU scenrio often advanced unified memory
is used, which can be multi channel for example.
The high L1 and L2 hit rate for shaders, translates
also to pi-WAM. Which shows, in the example I used,
a quite good GPU saturation, less memory bound and
more compute bound. I have compute these figures already.
Bye
Mild Shock schrieb:
Hi,
You are a moron right? Modern GPU has its
own L1,..,Ln caches. RTFM ASAP you brain
amputated vodka alcoholic.
Russian Hardbass - Crazy Dance
https://www.youtube.com/watch?v=6OoglRuT0jY
Bye
Jeiker Makulov schrieb:
Mild Shock wrote:find how
Yes of course, Micro Penis will know. Nothing, LoL But here we
RAM. Theyvillage idiot Micro Penis is.
The Cache Identity Crisis by Micro Penis:
Claiming that L1/L2/L3 caches and core topology are "embedded cpu
registers area,
it has nothing to do with the ram." That is a brilliant piece of
hardware fiction.
Caches are SRAM acting as a hierarchical
imbecile, formatting code like that puts you in prison in normal
countries. You are using entire lines for single chars, ie-a "}" etc, >>> -a> imbecile.
also, fucking lying illiterate imbecile
L1, L2, and L3 caches are located on the CPU, not in the system
form a high-speed memory hierarchy embedded directly within orimmediately
adjacent to the processor cores to minimize data access latency.
Hi,
Its actually quite amazing. Gemini, DeepSeek,
OpenAI all know Dmitriy V'jukov. I have asked
the IntelliJ integrated Freeium AI to generate
some code for me, I guess their service uses
by default OpenAI (Codex), and had it reviewed
by Gemini and DeepSeek. These AIs started lecturing
me about lazySet() in Java. But I went with set():
-a-a-a private static boolean enqueue(Queue q, Object data) {
-a-a-a-a-a-a-a int pos = q.enqueuePos.get();
-a-a-a-a-a-a-a for (; ; ) {
-a-a-a-a-a-a-a-a-a-a-a int index = pos & q.bufferMask;
-a-a-a-a-a-a-a-a-a-a-a int seq = q.sequences.get(index);
-a-a-a-a-a-a-a-a-a-a-a int dif = seq - pos;
-a-a-a-a-a-a-a-a-a-a-a if (dif == 0) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a if (q.enqueuePos.compareAndSet(pos, pos + 1)) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a q.data[index] = data;
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a q.sequences.set(index, pos + 1);
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a return true;
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a }
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a pos = q.enqueuePos.get();
-a-a-a-a-a-a-a-a-a-a-a } else if (dif < 0) {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a return false;
-a-a-a-a-a-a-a-a-a-a-a } else {
-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a pos = q.enqueuePos.get();
-a-a-a-a-a-a-a-a-a-a-a }
-a-a-a-a-a-a-a }
-a-a-a }
The above version seems to be more suitable
for my purpose, since it allows polling, it
basically implements offer(). While the
version posted on in the lock free group
by Chris M. Thomasson implements a spin wait
blocking put() already.
But I didn't port it yet to JavaScript or WGSL.
Currently busy with realizing a Worker facade
in JavaScript for a CPU backend that will run
in both the browser and node.js.
Bye
Chris M. Thomasson schrieb:
On 7/19/2026 2:04 PM, Chris M. Thomasson wrote:
I am friends with Dmitry Vyukov from way back, 23+ years ago.
Actually, I helped him find some bugs in Relacy when it was in
pre-alpha way back on comp.programming.threads. Also, I created some
neat eventcount algos, read all:
https://gist.github.com/mratsim/04a29bdd98d6295acda4d0677c4d0041
For starters... ;^)
also, I made a neat alteration to one of Dmitry Vyukov MPMC queues,
check it out:
https://groups.google.com/g/lock-free/c/acjQ3-89abE/m/a6-Di0GZsyEJ
can you get to the link? Thanks. Read all.
Mild Shock schrieb:
Hi,
I'm a spinner, I'm a sinner
I spin on CAS loops for my dinner
Some call it busy-wait, I call it fate
When the queue is empty, I just rotate
2 Producer threads, 2 Consumer threads
Each Producer generating 8192 items
the single common channel between
them 128 elements large.
public class DmitryVyukov
8 ms
public class DougLea
10 ms
On the Ryzen AI 7 350 which has 8 physical
cores with hyperthreading, making it 16
logical cores.
LoL
Bye
See also:
Producer-Consumer Queues by Dmitry Vyukov
https://sites.google.com/site/1024cores/home/lock-free-algorithms/queues
Steve Miller Band - The Joker (Official Music Video)
https://www.youtube.com/watch?v=dV3AziKTBUo
Mild Shock schrieb:
Hi,
Ok I was looking at this learning challenge,
producing vector (y1,y2,y3,y4) from a vector
(x1,x2,x3,x4), System R can do it via least square?
| 0 0 0 1 |-a-a | x1 |-a-a-a-a | x4 |
| 0 0 1 0 |-a-a | x2 |-a =-a | x3 |
| 0 1 0 0 |-a-a | x3 |-a-a-a-a | x2 |
| 1 0 0 0 |-a-a | x4 |-a-a-a-a | x1 |
How it started:
"multiplicative RNNs arises naturally from a
proof-theoretic interpretation of next-token
prediction as nested intuitionistic implication"
Paul Tarau - 2026
https://arxiv.org/abs/2601.19915
How its going:
"Dave uses a PDP-11 to train a real Neural
Network complete with Transformers and
Attention so you can see them at their most basic."
Mr. Taskmanager - 2026
https://www.youtube.com/watch?v=OUE3FSIk46g
We see Doctor Frankstein in action from
the Bronze Age of Computing, producing
a Humunkulus, the progenitor of todays
Bulgakov Shuriks in the Hyperscale Age!
Bye
P.S.: My impression neither cut to the core, that
this incredible transformer most likely
produced this deterministic attention:
| -1 | * | k | + | 5 | = | k' |
Or differently expressed y_k = x_{5-k}.
How did the transformer do it? It produced
a neural network with 1216 parameters, but
didn't use embeddings or polar encoding
of positions. But if we strip the noise
and denoise from the position encoding,
the denoise is done via softmax. We somehow
must get the above, right? I still need to
verify my claim! BTW: The PDP-11 assembly
from 1979 uses wider example not with n=4
but with n=8.
imageAtomicAdd is a damn good one for accumulation buffers.
Hi,--- Synchronet 3.22a-Linux NewsLink 1.2
I didn't find Futex in WebGPU / WGSL.
The website WebGPU fundamentals is on
GitHub. I did a search here:
https://github.com/webgpu/webgpufundamentals
In Java I can use Doug Leas queue.
In WebGPU / WGSL I will mostlikely
adopt Dmitry Vyukov , for a first stab.
Who is Doug lea?
He wrote Concurrent Programming in
Java: Design Principles and Patterns
https://en.wikipedia.org/wiki/Doug_Lea
He is behind most of the concurrency promitives
in Java. Including the array backed queue that
I tested. Remember the results I had:
public class DmitryVyukov
8 ms
public class DougLea
10 ms
Bye
On 07/20/2026 05:10 AM, Mild Shock wrote:..
Forget GPUs and forget you too.
Hi,
I am developing against WebGPU / WGSL.
An overview of WebGPU / WGSL is found here:
https://github.com/webgpu/webgpufundamentals
There is no imageAtomicAdd in WGSL.
imageAtomicAdd is from WebGL / GLSL.
These are two different things:
WebGPU / WGSL : Wrapper for Vulcan, Direct 12, or Metal
WebGL / GLSL : Wrapper for OpenGL
Chris M. Thomasson schrieb:
imageAtomicAdd is a damn good one for accumulation buffers.
Mild Shock schrieb:
Hi,
I didn't find Futex in WebGPU / WGSL.
The website WebGPU fundamentals is on
GitHub. I did a search here:
https://github.com/webgpu/webgpufundamentals
In Java I can use Doug Leas queue.
In WebGPU / WGSL I will mostlikely
adopt Dmitry Vyukov , for a first stab.
Who is Doug lea?
He wrote Concurrent Programming in
Java: Design Principles and Patterns
https://en.wikipedia.org/wiki/Doug_Lea
He is behind most of the concurrency promitives
in Java. Including the array backed queue that
I tested. Remember the results I had:
public class DmitryVyukov
8 ms
public class DougLea
10 ms
Bye
In comp.lang.prolog Ross Finlayson <ross.a.finlayson@gmail.com> wrote:
On 07/20/2026 05:10 AM, Mild Shock wrote:..
Forget GPUs and forget you too.
The joke ends "and the pig likes it".
Hi,
You guys are not paying attention. GPU is 10x
more performant than CPU for certain integerish
payload on the pi-WAM. 11.4 GLips on a GPU is
ca 10x more than 1.7 GLips on a CPU:
11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget
Parallel -C-WAM: 1.7 Giga Lips on a CPU
https://medium.com/2989/8a984e75af44
So GPUs are by way not dead. Especially since
they can be accessed via WebGPU / WGSL. But
the line between GPU and CPU increasingly
blurrs. It already happens for APUs in
that unified memory puts them into the same
RAM space. But the programming means differ.
Also unified memory doesn't mean that
the GPU sees memory the same way as a CPU.
To get a taste how a GPU sees memory:
Memory Model
-a-a-a Execution Barriers
-a-a-a Fence and Address Spaces
-a-a-a Memory Model GFX6-GFX9
-a-a-a Memory Model GFX90A
-a-a-a Memory Model GFX942
-a-a-a Memory Model GFX10-GFX11
-a-a-a Memory Model GFX12
-a-a-a Memory Model GFX125x
https://llvm.org/docs/AMDGPUUsage.html
WebGPU / WGSL seems to have good support,
since it can have Vulkan, Direct 12, or Metal
as GPU low level interface. Maybe more
platforms, something NVIDIA I guess. But I
don't know the details, how this is all done.
Bye
R Kym Horsell schrieb:
In comp.lang.prolog Ross Finlayson <ross.a.finlayson@gmail.com> wrote:
On 07/20/2026 05:10 AM, Mild Shock wrote:..
Forget GPUs and forget you too.
The joke ends "and the pig likes it".
Hi,
Concerning programmig means of GPUs,
"CUDA edge" refers to the massive competitive
advantage NVIDIA holds in artificial intelligence
due to its proprietary parallel computing platform,
CUDA (Compute Unified Device Architecture).
WebGPU and WGSL (WebGPU Shading Language)
represent a massive shift in AI because they
bring high-performance hardware acceleration
directly to the web browser without relying
on proprietary ecosystems like CUDA.
Developed by the W3C GPU for the Web
Community Group, this standard allows web
applications to interact directly with native
graphics APIs like Microsoft DirectX 12,
Apple Metal, and Vulkan
Bye
Some WebGPU show cases:
Frameworks like WebLLM leverage WebGPU to
run large language models locally inside
browsers like Google Chrome, providing
completely private, offline AI assistants.
Developers use libraries like Hugging Face
Transformers.js v3 to run computer vision,
speech recognition, and natural language
processing tasks directly on consumer
laptops and smartphones.
Mild Shock schrieb:
Hi,
You guys are not paying attention. GPU is 10x
more performant than CPU for certain integerish
payload on the pi-WAM. 11.4 GLips on a GPU is
ca 10x more than 1.7 GLips on a CPU:
11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget
Parallel -C-WAM: 1.7 Giga Lips on a CPU
https://medium.com/2989/8a984e75af44
So GPUs are by way not dead. Especially since
they can be accessed via WebGPU / WGSL. But
the line between GPU and CPU increasingly
blurrs. It already happens for APUs in
that unified memory puts them into the same
RAM space. But the programming means differ.
Also unified memory doesn't mean that
the GPU sees memory the same way as a CPU.
To get a taste how a GPU sees memory:
Memory Model
-a-a-a-a Execution Barriers
-a-a-a-a Fence and Address Spaces
-a-a-a-a Memory Model GFX6-GFX9
-a-a-a-a Memory Model GFX90A
-a-a-a-a Memory Model GFX942
-a-a-a-a Memory Model GFX10-GFX11
-a-a-a-a Memory Model GFX12
-a-a-a-a Memory Model GFX125x
https://llvm.org/docs/AMDGPUUsage.html
WebGPU / WGSL seems to have good support,
since it can have Vulkan, Direct 12, or Metal
as GPU low level interface. Maybe more
platforms, something NVIDIA I guess. But I
don't know the details, how this is all done.
Bye
R Kym Horsell schrieb:
In comp.lang.prolog Ross Finlayson <ross.a.finlayson@gmail.com> wrote:
On 07/20/2026 05:10 AM, Mild Shock wrote:..
Forget GPUs and forget you too.
The joke ends "and the pig likes it".
Hi,
I am developing against WebGPU / WGSL.
An overview of WebGPU / WGSL is found here:
https://github.com/webgpu/webgpufundamentals
There is no imageAtomicAdd in WGSL.
imageAtomicAdd is from WebGL / GLSL.
These are two different things:
WebGPU / WGSL : Wrapper for Vulcan, Direct 12, or Metal
WebGL / GLSL : Wrapper for OpenGL
Chris M. Thomasson schrieb:
imageAtomicAdd is a damn good one for accumulation buffers.
Mild Shock schrieb:
Hi,
I didn't find Futex in WebGPU / WGSL.
The website WebGPU fundamentals is on
GitHub. I did a search here:
https://github.com/webgpu/webgpufundamentals
In Java I can use Doug Leas queue.
In WebGPU / WGSL I will mostlikely
adopt Dmitry Vyukov , for a first stab.
Who is Doug lea?
He wrote Concurrent Programming in
Java: Design Principles and Patterns
https://en.wikipedia.org/wiki/Doug_Lea
He is behind most of the concurrency promitives
in Java. Including the array backed queue that
I tested. Remember the results I had:
public class DmitryVyukov
8 ms
public class DougLea
10 ms
Bye
Hi,
But I am nowhere using WebGL / GLSL.
The experiment here GPU versus CPU,
was done with WebGPU / WGSL:
11.4 Giga Lips with a Budget Laptop https://github.com/Jean-Luc-Picard-2021/gigabudget
Parallel -C-WAM: 1.7 Giga Lips on a CPU
https://medium.com/2989/8a984e75af44
I do not intend to redo the experiment
"gigabudget" with WebGL / GSLS. It would
appear to me like flogging a dead horse,
a technology that has reached EOL, namely
OpenGL which is in the phase of end of lifetime.
Bye
Mild Shock schrieb:
Hi,
I am developing against WebGPU / WGSL.
An overview of WebGPU / WGSL is found here:
https://github.com/webgpu/webgpufundamentals
There is no imageAtomicAdd in WGSL.
imageAtomicAdd is from WebGL / GLSL.
These are two different things:
WebGPU / WGSL : Wrapper for Vulcan, Direct 12, or Metal
WebGL / GLSL : Wrapper for OpenGL
Chris M. Thomasson schrieb:
imageAtomicAdd is a damn good one for accumulation buffers.
Mild Shock schrieb:
Hi,
I didn't find Futex in WebGPU / WGSL.
The website WebGPU fundamentals is on
GitHub. I did a search here:
https://github.com/webgpu/webgpufundamentals
In Java I can use Doug Leas queue.
In WebGPU / WGSL I will mostlikely
adopt Dmitry Vyukov , for a first stab.
Who is Doug lea?
He wrote Concurrent Programming in
Java: Design Principles and Patterns
https://en.wikipedia.org/wiki/Doug_Lea
He is behind most of the concurrency promitives
in Java. Including the array backed queue that
I tested. Remember the results I had:
public class DmitryVyukov
8 ms
public class DougLea
10 ms
Bye
Hi,
imageAtomicAdd is trivial, but it does
not help with bounded buffers. I already
did imageAtomicAdd, in an experiment,
where pi-WAM implemented an in and out
buffer as follows in Java, which can
be trivially ported to WebGPU / WGSL,
by using atomic(i32) and AtomicAdd:
-a-a private static final class PiChan {
-a-a-a-a-a-a-a private int[] buf;
-a-a-a-a-a-a-a private AtomicInteger pos;
-a-a-a }
-a-a-a private static final class PiWam {
-a-a-a-a-a-a-a private PiChan in;
-a-a-a Etc...
-a-a-a }
-a-a case 5: /* in */
-a-a-a-a-a-a-a int at = pi.in.pos.getAndAdd(obj);
-a-a-a-a-a-a-a for (int i = 0; i < obj; i++)
-a-a-a-a-a-a-a-a-a-a-a-a pi.state[offset + i] = pi.in.buf[at + i];
-a-a-a-a-a-a-a return 0;
But this is not the same like Dmitry
Vyukov buffer. Which has a maximum
capacity, and fails to go beyond this
capacity filling a buffer by a producer,
before a consumer made the buffer not
full again. My requirement for pi-WAM
are bounded buffers with a finite capacity.
Bye
Mild Shock schrieb:
Hi,
But I am nowhere using WebGL / GLSL.
The experiment here GPU versus CPU,
was done with WebGPU / WGSL:
11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget
Parallel -C-WAM: 1.7 Giga Lips on a CPU
https://medium.com/2989/8a984e75af44
I do not intend to redo the experiment
"gigabudget" with WebGL / GSLS. It would
appear to me like flogging a dead horse,
a technology that has reached EOL, namely
OpenGL which is in the phase of end of lifetime.
Bye
Mild Shock schrieb:
Hi,
I am developing against WebGPU / WGSL.
An overview of WebGPU / WGSL is found here:
https://github.com/webgpu/webgpufundamentals
There is no imageAtomicAdd in WGSL.
imageAtomicAdd is from WebGL / GLSL.
These are two different things:
WebGPU / WGSL : Wrapper for Vulcan, Direct 12, or Metal
WebGL / GLSL : Wrapper for OpenGL
Chris M. Thomasson schrieb:
imageAtomicAdd is a damn good one for accumulation buffers.
Mild Shock schrieb:
Hi,
I didn't find Futex in WebGPU / WGSL.
The website WebGPU fundamentals is on
GitHub. I did a search here:
https://github.com/webgpu/webgpufundamentals
In Java I can use Doug Leas queue.
In WebGPU / WGSL I will mostlikely
adopt Dmitry Vyukov , for a first stab.
Who is Doug lea?
He wrote Concurrent Programming in
Java: Design Principles and Patterns
https://en.wikipedia.org/wiki/Doug_Lea
He is behind most of the concurrency promitives
in Java. Including the array backed queue that
I tested. Remember the results I had:
public class DmitryVyukov
8 ms
public class DougLea
10 ms
Bye
Hi,
I am still singing this song:
"I'm a spinner, I'm a sinner
I spin on CAS loops for my dinner
Some call it busy-wait, I call it fate
When the queue is empty, I just rotate"
In Dmitry Vyukov multiple producer and
multiple consuer, the assumption is
that the capacity is a multiple power
of 2. This way some inveriants hold
computing seq - pos, even of the i32
arithmetc overflows, the difference
will still be in the interval -capacity
... capacity , as the AI chat bot explained
me. The initialization of a Dmitry Vyukov
then doesn't store capacity itself, but
a mask derived from capacity:
-a-a-a private static void init(Queue q, int size) {
-a-a-a-a-a-a-a q.bufferMask = size - 1;
-a-a-a-a-a-a-a q.sequences = new AtomicIntegerArray(size);
-a-a-a-a-a-a-a for (int i = 0; i < size; i++)
-a-a-a-a-a-a-a-a-a-a-a q.sequences.set(i, i);
-a-a-a-a-a-a-a q.data = new Object[size];
-a-a-a-a-a-a-a q.enqueuePos = new AtomicInteger(0);
-a-a-a-a-a-a-a q.dequeuePos = new AtomicInteger(0);
-a-a-a }
I cannot use imageAtomicAdd, which wouldn't
have a finite capacity. But as you see
I have already a prototype of a Queue
with a finite capacity. And the results
for a systolic system are quite good:
public class DmitryVyukov
8 ms
public class DougLea
10 ms
Have Fun!
Bye
Mild Shock schrieb:
Hi,
imageAtomicAdd is trivial, but it does
not help with bounded buffers. I already
did imageAtomicAdd, in an experiment,
where pi-WAM implemented an in and out
buffer as follows in Java, which can
be trivially ported to WebGPU / WGSL,
by using atomic(i32) and AtomicAdd:
-a-a-a private static final class PiChan {
-a-a-a-a-a-a-a-a private int[] buf;
-a-a-a-a-a-a-a-a private AtomicInteger pos;
-a-a-a-a }
-a-a-a-a private static final class PiWam {
-a-a-a-a-a-a-a-a private PiChan in;
-a-a-a-a Etc...
-a-a-a-a }
-a-a-a case 5: /* in */
-a-a-a-a-a-a-a-a int at = pi.in.pos.getAndAdd(obj);
-a-a-a-a-a-a-a-a for (int i = 0; i < obj; i++)
-a-a-a-a-a-a-a-a-a-a-a-a-a pi.state[offset + i] = pi.in.buf[at + i];
-a-a-a-a-a-a-a-a return 0;
But this is not the same like Dmitry
Vyukov buffer. Which has a maximum
capacity, and fails to go beyond this
capacity filling a buffer by a producer,
before a consumer made the buffer not
full again. My requirement for pi-WAM
are bounded buffers with a finite capacity.
Bye
Mild Shock schrieb:
Hi,
But I am nowhere using WebGL / GLSL.
The experiment here GPU versus CPU,
was done with WebGPU / WGSL:
11.4 Giga Lips with a Budget Laptop
https://github.com/Jean-Luc-Picard-2021/gigabudget
Parallel -C-WAM: 1.7 Giga Lips on a CPU
https://medium.com/2989/8a984e75af44
I do not intend to redo the experiment
"gigabudget" with WebGL / GSLS. It would
appear to me like flogging a dead horse,
a technology that has reached EOL, namely
OpenGL which is in the phase of end of lifetime.
Bye
Mild Shock schrieb:
Hi,
I am developing against WebGPU / WGSL.
An overview of WebGPU / WGSL is found here:
https://github.com/webgpu/webgpufundamentals
There is no imageAtomicAdd in WGSL.
imageAtomicAdd is from WebGL / GLSL.
These are two different things:
WebGPU / WGSL : Wrapper for Vulcan, Direct 12, or Metal
WebGL / GLSL : Wrapper for OpenGL
Chris M. Thomasson schrieb:
imageAtomicAdd is a damn good one for accumulation buffers.
Mild Shock schrieb:
Hi,
I didn't find Futex in WebGPU / WGSL.
The website WebGPU fundamentals is on
GitHub. I did a search here:
https://github.com/webgpu/webgpufundamentals
In Java I can use Doug Leas queue.
In WebGPU / WGSL I will mostlikely
adopt Dmitry Vyukov , for a first stab.
Who is Doug lea?
He wrote Concurrent Programming in
Java: Design Principles and Patterns
https://en.wikipedia.org/wiki/Doug_Lea
He is behind most of the concurrency promitives
in Java. Including the array backed queue that
I tested. Remember the results I had:
public class DmitryVyukov
8 ms
public class DougLea
10 ms
Bye
On 07/20/2026 05:10 AM, Mild Shock wrote:
Hi,
If you don't know how GPU caches work.
Just RTFM the RDNA 3.5 specs! They have
some explanations in the docs, what jobs
the GPU cache lines do, in relation to
what instructions:
Abbreviation for rCyRead The Fucking ManualrCO.
http://www.catb.org/esr/jargon/html/R/RTFM.html
Here some RDNA 4.0 specs (smaller GPUs):
"RDNA4" Instruction Set Architecture
Reference Guide - 7-April-2025
https://docs.amd.com/v/u/en-US/rdna4-instruction-set-architecture
Here some CDNA 4.0 specs (bigger GPUs):
CDNA4 Instruction Set Architecture
Reference Guide - 5-August-2025
https://www.amd.com/content/dam/amd/en/documents/instinct-tech-docs/instruction-set-architectures/amd-instinct-cdna4-instruction-set-architecture.pdf
Bye
But the difference between CDNA GPUs and
RDNA GPUs seems not to be that big, at least
in 2025:
Commitment to High-Performance
Computing in the Age of AI
https://docs.amd.com/v/u/en-US/precision-at-scale-amd-whitepaper
While the Copilot+ threshold is 45 TFLOPs,
a MI355X, Datacenter, rack-mounted, liquid cooling,
has surely more memory, but only 157 TFLOPS.
And RTX 5090, Desktop, 600W+ TDP, massive cooling,
has also only 100 TFLOPS. So it looks these
Copilot+ AI Laptops are pretty swell, arent they?
Mild Shock schrieb:
Hi,
completely nonsense. Not even correct grammatically
Yes of course, Micro Penis will know. Nothing, LoL
But here we find how village idiot Micro Penis is.
The Cache Identity Crisis by Micro Penis:
Claiming that L1/L2/L3 caches and core
topology are "embedded cpu registers area,
it has nothing to do with the ram." That is
a brilliant piece of hardware fiction.
Caches are SRAM acting as a hierarchical
staging ground for and backed by system
or device RAM, not an alternate dimension
divorced from memory entirely.
And calling cache hierarchies "registers"
is like confusing a warehouse parking lot
with the glove compartment of a single car.
Never heard of the crossbar in AMD GPUs?
LoL
Bye
Obelin Baisaroff schrieb:
Ross Finlayson wrote:
access to memory, has that these days with
L1/L2/L3 caches and the proximity and affinity
in the topology of the cores and
those are embedded cpu registers area, it has
nothing to do with the ram. Works by higher
clocks compared, but merely useless in AI.
Romelio Balakhonsky schrieb:
Mild Shock wrote:pos +
me about lazySet() in Java. But I went with set():
private static boolean enqueue(Queue q, Object data) {
int pos = q.enqueuePos.get();
for (; ; ) {
int index = pos & q.bufferMask; int seq =
q.sequences.get(index);
int dif = seq - pos;
if (dif == 0) {
if (q.enqueuePos.compareAndSet(pos, pos + 1)) {
q.data[index] = data; q.sequences.set(index,
1);
return true;
}
pos = q.enqueuePos.get();
} else if (dif < 0) {
return false;
} else {
pos = q.enqueuePos.get();
}
}
}
The above version seems to be more suitable for my purpose, since it >>> >> allows polling, it basically implements offer(). While the
completely nonsense. Not even correct grammatically
Forget GPUs and forget you too.
On 7/20/2026 3:15 PM, Mild Shock wrote:
[...]
Have you ever used imageAtomicAdd before?
Also forget your spammy sock-puppets
and the grimy, gropy mitts up in them.
Hi,
I went to holdays in June 2026, had an idea for
a pi-WAM based on a Hack, the later is described here:
Emulating -C-WAM in Dogelog Player
https://medium.com/2989/de9cd29c7d37
The Elements of Computing Systems
https://mitpress.mit.edu/9780262539807
In July 2026 I did the CPU and GPU experiments,
moving from emulator to native executor based in
realizing Hack as a concrete virtual machine,
and not as an abstract machine emulated in Prolog.
The GPU experiments were done in WebGPU / WGSL.
So no, I never used OpenGL Version 4.2 and later.
Also the name imageAtomicAdd indicates that it
imageAtomicAdd is rather from a render shader,
while my GPU experiment uses a compute shader.
Es specially I need GPU compute shaders, which
are not executed in lock step, but rather have
indepdendent thread state, also known as MIMD.
"In computing, multiple instruction, multiple
data (MIMD) is a technique employed to
achieve parallelism. " https://en.wikipedia.org/wiki/Multiple_instruction,_multiple_data
MIMID showed up 2017 with NVIDIA Volta cards.
But is now realized by Intel Arc, Snapdragon Adreno
and AMD RDNA as well.
Bye
Chris M. Thomasson schrieb:
On 7/20/2026 3:15 PM, Mild Shock wrote:
[...]
Have you ever used imageAtomicAdd before?
Hi,
What sock puppets? Just look at the headers
of the source of my posts, its always:
From: Mild Shock <janburse@fastmail.fm>
Did corona burn out the last of your braincells,
and now rossy boy is going paranoid.
Or what is the meaning of your herpes blister
gibberish that you are regularly posting?
Bye
Ross Finlayson schrieb:
Also forget your spammy sock-puppets
and the grimy, gropy mitts up in them.
Hi,
------------------- begin --------------------
Teaching Micro Penis Vilage Idiot
------------------- begin --------------------
You dont have to use WebLLM, respectively
WebGPU / WGSL literally, just read the next
post I did AND use your brains moron:
Like WebAssembly before it, WebGPU has "escaped" the browser
http://localhost:567921/ is private you moron.,
or what ever port REST is using. You typically access
an offline AI assistant, via some REST end-point
on your machine. Nothing to do with Google Chrome
browser security. You can make it as private as you want, by
having a firewall and not outward or inward
connection at all, only your REST end-point
on your machine. Or if you want a REST end-point
on a server of yours in the same intranet.
You don't need to use the internet, or put
something on the extranet, or use some sort of
subscription. What you need is access through
the firewall to download the REST software
and the LLM model. Tools like LM Studio and oMLX
offer this download and also install REST endpoint.
------------------- end --------------------
Teaching Micro Penis Vilage Idiot}
------------------- end --------------------
Bye
Will Bakshandaev schrieb:
Mild Shock wrote:
Frameworks like WebLLM leverage WebGPU to run large language models
locally inside browsers like Google Chrome, providing completely
private, offline AI assistants.
since when google chrome private, think again
Windows Recall takes a screenshot of a user's desktop every fewseconds, then uses on-device large language models to allow a user to
Hi,
Currently companies such as Apple, Windows, etc..
are hardning their operating systems, so
that they can provide agentic AI sandboxes.
Problem is an agentic AI, that acts on your
behalf, when not enough supervised, might
do all kind of stuff on its own. So how do you
have harder borders. Besides companies that
write operating systems, there is also a cottage
industry now that adresses this paranoia,
here an example from a former Prologer:
Stop guessing what your coding agent just did
Prempti: Guardrails and Observability for AI Coding Agents. https://prempti.falco.org/
IntelliJ doesn't have this problem, it shows a
not yet hyper locally commited change, in the editor,
created by the AI, that you can review, and
then hyper locally commit in the editor. Only then
it lands in the file system. But also there it
will be subject to the local history and repository
version system. So the IntelliJ AI is pretty smartly
implemented, and hooks into their editors and newly
introduced hyper change visualization, a feature that
probably codemirror doesn't have yet. Have to double check.
Have Fun!
Bye
Mild Shock schrieb:
Hi,
------------------- begin --------------------
Teaching Micro Penis Vilage Idiot
------------------- begin --------------------
You dont have to use WebLLM, respectively
WebGPU / WGSL literally, just read the next
post I did AND use your brains moron:
Like WebAssembly before it, WebGPU has "escaped" the browser
https://blog.4dpipeline.com/client-side-ai-is-here-how-webgpu-transforms-your-gpu-server-economics
http://localhost:567921/ is private you moron.,
or what ever port REST is using. You typically access
an offline AI assistant, via some REST end-point
on your machine. Nothing to do with Google Chrome
browser security. You can make it as private as you want, by
having a firewall and not outward or inward
connection at all, only your REST end-point
on your machine. Or if you want a REST end-point
on a server of yours in the same intranet.
You don't need to use the internet, or put
something on the extranet, or use some sort of
subscription. What you need is access through
the firewall to download the REST software
and the LLM model. Tools like LM Studio and oMLX
offer this download and also install REST endpoint.
------------------- end --------------------
Teaching Micro Penis Vilage Idiot}
------------------- end --------------------
Bye
Will Bakshandaev schrieb:
Mild Shock wrote:
Frameworks like WebLLM leverage WebGPU to run large language models
locally inside browsers like Google Chrome, providing completely
private, offline AI assistants.
since when google chrome private, think again
Mild Shock wrote:
http://localhost:567921/ is private you moron.,access an offline AI assistant, via some REST end-point
or what ever port REST is using. You typically
yet one more proof this half german inbreed
is an imbecile, ports go up to 16bits/64k only,
idiot, you cant have a localhost: whatever
wrong number you put there. You extreme fucking idiot.
on your machine. Nothing to do with Google
Chrome browser security. You can make it as
private as you want, by having a firewall
and not outward or inward
yes, i can see your point, they just want
your private cellphone number, there rest
is private and free, idiot
Hi,
What would an EMACs guru say. Can
EMACs process the equivalent of these
HTML tags:
The <ins> HTML element represents a
range of text that has been added to a document. https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/Elements/ins
The <del> HTML element represents a range
of text that has been deleted from a document. https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/Elements/del
This can be used when rendering "track changes"
or source code diff information, for example.
Office word can do it as well, track changes.
In principle somehow, every editor that has undo
and redo, dunno, can Excel show changes? What was
Windows Recall again on an AI Laptop?
Windows Recall takes a screenshot of a user's desktop every fewseconds, then uses on-device large language models to allow a user to retrieve items and information that had previously been on their screen. https://en.wikipedia.org/wiki/Windows_Recall
A little bit unstructured, compared to the
HTML tags and quite neurotic approach.
Bye
Mild Shock schrieb:
Hi,
Currently companies such as Apple, Windows, etc..
are hardning their operating systems, so
that they can provide agentic AI sandboxes.
Problem is an agentic AI, that acts on your
behalf, when not enough supervised, might
do all kind of stuff on its own. So how do you
have harder borders. Besides companies that
write operating systems, there is also a cottage
industry now that adresses this paranoia,
here an example from a former Prologer:
Stop guessing what your coding agent just did
Prempti: Guardrails and Observability for AI Coding Agents.
https://prempti.falco.org/
IntelliJ doesn't have this problem, it shows a
not yet hyper locally commited change, in the editor,
created by the AI, that you can review, and
then hyper locally commit in the editor. Only then
it lands in the file system. But also there it
will be subject to the local history and repository
version system. So the IntelliJ AI is pretty smartly
implemented, and hooks into their editors and newly
introduced hyper change visualization, a feature that
probably codemirror doesn't have yet. Have to double check.
Have Fun!
Bye
Mild Shock schrieb:
Hi,https://blog.4dpipeline.com/client-side-ai-is-here-how-webgpu-transforms-your-gpu-server-economics
------------------- begin --------------------
Teaching Micro Penis Vilage Idiot
------------------- begin --------------------
You dont have to use WebLLM, respectively
WebGPU / WGSL literally, just read the next
post I did AND use your brains moron:
Like WebAssembly before it, WebGPU has "escaped" the browser
http://localhost:567921/ is private you moron.,
or what ever port REST is using. You typically access
an offline AI assistant, via some REST end-point
on your machine. Nothing to do with Google Chrome
browser security. You can make it as private as you want, by
having a firewall and not outward or inward
connection at all, only your REST end-point
on your machine. Or if you want a REST end-point
on a server of yours in the same intranet.
You don't need to use the internet, or put
something on the extranet, or use some sort of
subscription. What you need is access through
the firewall to download the REST software
and the LLM model. Tools like LM Studio and oMLX
offer this download and also install REST endpoint.
------------------- end --------------------
Teaching Micro Penis Vilage Idiot}
------------------- end --------------------
Bye
Will Bakshandaev schrieb:
Mild Shock wrote:
Frameworks like WebLLM leverage WebGPU to run large language models
locally inside browsers like Google Chrome, providing completely
private, offline AI assistants.
since when google chrome private, think again
Hi,
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN" http://localhost:1234/api/v0/models
Response format
{
-a "object": "list",
-a "data": [
-a-a-a {
-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a "object": "model",
-a-a-a-a-a "type": "vlm",
-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
Mild Shock wrote:
http://localhost:567921/ is private you moron.,access an offline AI assistant, via some REST end-point
or what ever port REST is using. You typically
yet one more proof this half german inbreed
is an imbecile, ports go up to 16bits/64k only,
idiot, you cant have a localhost: whatever
wrong number you put there. You extreme fucking idiot.
on your machine. Nothing to do with Google
Chrome browser security. You can make it as
private as you want, by having a firewall
and not outward or inward
yes, i can see your point, they just want
your private cellphone number, there rest
is private and free, idiot
Mild Shock schrieb:
Hi,
What would an EMACs guru say. Can
EMACs process the equivalent of these
HTML tags:
The <ins> HTML element represents a
range of text that has been added to a document.
https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/Elements/ins
The <del> HTML element represents a range
of text that has been deleted from a document.
https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/Elements/del
This can be used when rendering "track changes"
or source code diff information, for example.
Office word can do it as well, track changes.
In principle somehow, every editor that has undo
and redo, dunno, can Excel show changes? What was
Windows Recall again on an AI Laptop?
Windows Recall takes a screenshot of a user's desktop every fewseconds, then uses on-device large language models to allow a user to
retrieve items and information that had previously been on their screen.
https://en.wikipedia.org/wiki/Windows_Recall
A little bit unstructured, compared to the
HTML tags and quite neurotic approach.
Bye
Mild Shock schrieb:
Hi,
Currently companies such as Apple, Windows, etc..
are hardning their operating systems, so
that they can provide agentic AI sandboxes.
Problem is an agentic AI, that acts on your
behalf, when not enough supervised, might
do all kind of stuff on its own. So how do you
have harder borders. Besides companies that
write operating systems, there is also a cottage
industry now that adresses this paranoia,
here an example from a former Prologer:
Stop guessing what your coding agent just did
Prempti: Guardrails and Observability for AI Coding Agents.
https://prempti.falco.org/
IntelliJ doesn't have this problem, it shows a
not yet hyper locally commited change, in the editor,
created by the AI, that you can review, and
then hyper locally commit in the editor. Only then
it lands in the file system. But also there it
will be subject to the local history and repository
version system. So the IntelliJ AI is pretty smartly
implemented, and hooks into their editors and newly
introduced hyper change visualization, a feature that
probably codemirror doesn't have yet. Have to double check.
Have Fun!
Bye
Mild Shock schrieb:
Hi,https://blog.4dpipeline.com/client-side-ai-is-here-how-webgpu-transforms-your-gpu-server-economics
------------------- begin --------------------
Teaching Micro Penis Vilage Idiot
------------------- begin --------------------
You dont have to use WebLLM, respectively
WebGPU / WGSL literally, just read the next
post I did AND use your brains moron:
Like WebAssembly before it, WebGPU has "escaped" the browser
http://localhost:567921/ is private you moron.,
or what ever port REST is using. You typically access
an offline AI assistant, via some REST end-point
on your machine. Nothing to do with Google Chrome
browser security. You can make it as private as you want, by
having a firewall and not outward or inward
connection at all, only your REST end-point
on your machine. Or if you want a REST end-point
on a server of yours in the same intranet.
You don't need to use the internet, or put
something on the extranet, or use some sort of
subscription. What you need is access through
the firewall to download the REST software
and the LLM model. Tools like LM Studio and oMLX
offer this download and also install REST endpoint.
------------------- end --------------------
Teaching Micro Penis Vilage Idiot}
------------------- end --------------------
Bye
Will Bakshandaev schrieb:
Mild Shock wrote:
Frameworks like WebLLM leverage WebGPU to run large language models >>> -a>>> locally inside browsers like Google Chrome, providing completely
private, offline AI assistants.
since when google chrome private, think again
Hi,
How confused is tiny winy penis?
For the 100th time the budget here:
11.4 Giga Lips with a Budget Laptop
Ryzen AI 7 350 w/ Radeon 860M https://github.com/Jean-Luc-Picard-2021/gigabudget
is a laptop and not a smartphone. It
has no cellphone number. And w/ means
integrated GPU on the silicon chip,
and not a GPU connected to the mainboard
via some PCI bus. The model is a acer
swift go, I already posted this info:
Swift Go 16 AI SFG16-61-R21J Notebook https://www.acer.com/ch-de/laptops/swift/swift-go-16-ai-amd/pdp/NX.JCREZ.007
Bye
Mild Shock schrieb:
Hi,
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN"
http://localhost:1234/api/v0/models
Response format
{
-a-a "object": "list",
-a-a "data": [
-a-a-a-a {
-a-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a-a "object": "model",
-a-a-a-a-a-a "type": "vlm",
-a-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
Mild Shock wrote:
http://localhost:567921/ is private you moron.,access an offline AI assistant, via some REST end-point
or what ever port REST is using. You typically
yet one more proof this half german inbreed
is an imbecile, ports go up to 16bits/64k only,
idiot, you cant have a localhost: whatever
wrong number you put there. You extreme fucking idiot.
on your machine. Nothing to do with Google
Chrome browser security. You can make it as
private as you want, by having a firewall
and not outward or inward
yes, i can see your point, they just want
your private cellphone number, there rest
is private and free, idiot
Mild Shock schrieb:
Hi,
What would an EMACs guru say. Can
EMACs process the equivalent of these
HTML tags:
The <ins> HTML element represents a
range of text that has been added to a document.
https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/Elements/ins >>>
The <del> HTML element represents a range
of text that has been deleted from a document.
https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/Elements/del >>>
This can be used when rendering "track changes"
or source code diff information, for example.
Office word can do it as well, track changes.
In principle somehow, every editor that has undo
and redo, dunno, can Excel show changes? What was
Windows Recall again on an AI Laptop?
Windows Recall takes a screenshot of a user's desktop every fewseconds, then uses on-device large language models to allow a user to
retrieve items and information that had previously been on their screen. >>> https://en.wikipedia.org/wiki/Windows_Recall
A little bit unstructured, compared to the
HTML tags and quite neurotic approach.
Bye
Mild Shock schrieb:
Hi,
Currently companies such as Apple, Windows, etc..
are hardning their operating systems, so
that they can provide agentic AI sandboxes.
Problem is an agentic AI, that acts on your
behalf, when not enough supervised, might
do all kind of stuff on its own. So how do you
have harder borders. Besides companies that
write operating systems, there is also a cottage
industry now that adresses this paranoia,
here an example from a former Prologer:
Stop guessing what your coding agent just did
Prempti: Guardrails and Observability for AI Coding Agents.
https://prempti.falco.org/
IntelliJ doesn't have this problem, it shows a
not yet hyper locally commited change, in the editor,
created by the AI, that you can review, and
then hyper locally commit in the editor. Only then
it lands in the file system. But also there it
will be subject to the local history and repository
version system. So the IntelliJ AI is pretty smartly
implemented, and hooks into their editors and newly
introduced hyper change visualization, a feature that
probably codemirror doesn't have yet. Have to double check.
Have Fun!
Bye
Mild Shock schrieb:
Hi,https://blog.4dpipeline.com/client-side-ai-is-here-how-webgpu-transforms-your-gpu-server-economics
------------------- begin --------------------
Teaching Micro Penis Vilage Idiot
------------------- begin --------------------
You dont have to use WebLLM, respectively
WebGPU / WGSL literally, just read the next
post I did AND use your brains moron:
Like WebAssembly before it, WebGPU has "escaped" the browser
models
http://localhost:567921/ is private you moron.,
or what ever port REST is using. You typically access
an offline AI assistant, via some REST end-point
on your machine. Nothing to do with Google Chrome
browser security. You can make it as private as you want, by
having a firewall and not outward or inward
connection at all, only your REST end-point
on your machine. Or if you want a REST end-point
on a server of yours in the same intranet.
You don't need to use the internet, or put
something on the extranet, or use some sort of
subscription. What you need is access through
the firewall to download the REST software
and the LLM model. Tools like LM Studio and oMLX
offer this download and also install REST endpoint.
------------------- end --------------------
Teaching Micro Penis Vilage Idiot}
------------------- end --------------------
Bye
Will Bakshandaev schrieb:
Mild Shock wrote:
Frameworks like WebLLM leverage WebGPU to run large language
locally inside browsers like Google Chrome, providing completely >>>> -a>>> private, offline AI assistants.
since when google chrome private, think again
you stupid half german, the comparisons along
memories arrays gpu card located, are taking
place parallel without cpu intervention, you
fucking illiterate idiot. You soon will become a
quarter german hence 3/4 russian old days,
historically. It's coming
German energy crisis caused by rCylack of Russian gasrCO rCo Merz https://www.rt.com/news/643254-germany-crisis-russian-gas/
Mild Shock wrote:
Maybe you could post some subtantial critique moron? Instead of
gibberish all the time. What does a cellphone number have to do with a
REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
you lying bitch, your arse is burning, you said google chrome security,
and gave them your cellphone number, fucking idiot. And you put wrong
ports numbers along the localhost: idiot
Hi,--- Synchronet 3.22a-Linux NewsLink 1.2
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN" http://localhost:1234/api/v0/models
Response format
{
-a "object": "list",
-a "data": [
-a-a-a {
-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a "object": "model",
-a-a-a-a-a "type": "vlm",
-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
Hi,
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
secure sand box, for executing JavaScript in the main
and in isolated siloed workers. Also you seem not to see
that I post CHF price tags and CHF product listings.
So what makes you think Switzerland == Germany,
when you rant like below. Even a dog knows that a
non-EU country cannot be the same as EU country:
you stupid half german, the comparisons along
memories arrays gpu card located, are taking
place parallel without cpu intervention, you
fucking illiterate idiot. You soon will become a
quarter german hence 3/4 russian old days,
historically. It's coming
German energy crisis caused by rCylack of Russian gasrCO rCo Merz https://www.rt.com/news/643254-germany-crisis-russian-gas/
Bye
P.S.: EU = European Union
Audie Balaban schrieb:
Mild Shock wrote:
Maybe you could post some subtantial critique moron? Instead of
gibberish all the time. What does a cellphone number have to do with a
REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
you lying bitch, your arse is burning, you said google chrome security, and gave them your cellphone number, fucking idiot. And you put wrong ports numbers along the localhost: idiot
Mild Shock schrieb:
Hi,
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN"
http://localhost:1234/api/v0/models
Response format
{
-a-a "object": "list",
-a-a "data": [
-a-a-a-a {
-a-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a-a "object": "model",
-a-a-a-a-a-a "type": "vlm",
-a-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
Hi,
But I already wrote like a 100times that
you can use WebGPU outside of a browser.
The same applies for Workers.
You can use WebGPU and Workers from within
node.js. No chrome browser (from google)
needed, or safari (from apple) or edge (from
microsoft). If you don't need a User Experience
(UX), i.e. if its enough to go headless with
your use case, for example implementing a
REST endpoint, you can use node.js for both
WebGPU and Workers. I posted the WebGPU headless
article, like 100times already:
Like WebAssembly before it, WebGPU has "escaped" the browser https://blog.4dpipeline.com/client-side-ai-is-here-how-webgpu-transforms-your-gpu-server-economics
I also recently blogged about how to
use multiple workers from a main via browser
AND node.js. It not extremly difficult:
Parallel -C-WAM: JavaScript Workers as CPU Backend https://medium.com/2989/5ef903e5e785
But well your Syphilis Brain has obviously
shut down its operation, silly Micro Penis.
Bye
Mild Shock schrieb:
Hi,
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
secure sand box, for executing JavaScript in the main
and in isolated siloed workers. Also you seem not to see
that I post CHF price tags and CHF product listings.
So what makes you think Switzerland == Germany,
when you rant like below. Even a dog knows that a
non-EU country cannot be the same as EU country:
you stupid half german, the comparisons along
memories arrays gpu card located, are taking
place parallel without cpu intervention, you
fucking illiterate idiot. You soon will become a
quarter german hence 3/4 russian old days,
historically. It's coming
German energy crisis caused by rCylack of Russian gasrCO rCo Merz
https://www.rt.com/news/643254-germany-crisis-russian-gas/
Bye
P.S.: EU = European Union
Audie Balaban schrieb:
Mild Shock wrote:with a
Maybe you could post some subtantial critique moron? Instead of
gibberish all the time. What does a cellphone number have to do
security,REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
you lying bitch, your arse is burning, you said google chrome
and gave them your cellphone number, fucking idiot. And you put wrong
ports numbers along the localhost: idiot
Mild Shock schrieb:
Hi,
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN"
http://localhost:1234/api/v0/models
Response format
{
-a-a "object": "list",
-a-a "data": [
-a-a-a-a {
-a-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a-a "object": "model",
-a-a-a-a-a-a "type": "vlm",
-a-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
Hi,
Micro Penis, you possible still think when I wrote
mobile grade GPU, thats it is a GPU only for
smallphones, right?
Maybe stop your hobby interpretation of everything,
and RTFM . And try to get a professional understanding .
I admit it might take you a few months to
Keep up with the Kardashians so to speek, i.e. the
reality show of GPUs, and not what your Syphilis Brain
halucinates. To get the basic understanding
that is needed to follow what I post.
Good Luck!
Bye
RTFM = Abbreviation for rCyRead The Fucking ManualrCO. http://www.catb.org/esr/jargon/html/R/RTFM.html
Keeping Up with the Kardashians https://en.wikipedia.org/wiki/Keeping_Up_with_the_Kardashians
Mild Shock schrieb:
Hi,
But I already wrote like a 100times that
you can use WebGPU outside of a browser.
The same applies for Workers.
You can use WebGPU and Workers from within
node.js. No chrome browser (from google)
needed, or safari (from apple) or edge (from
microsoft). If you don't need a User Experience
(UX), i.e. if its enough to go headless with
your use case, for example implementing a
REST endpoint, you can use node.js for both
WebGPU and Workers. I posted the WebGPU headless
article, like 100times already:
Like WebAssembly before it, WebGPU has "escaped" the browser
https://blog.4dpipeline.com/client-side-ai-is-here-how-webgpu-transforms-your-gpu-server-economics
I also recently blogged about how to
use multiple workers from a main via browser
AND node.js. It not extremly difficult:
Parallel -C-WAM: JavaScript Workers as CPU Backend
https://medium.com/2989/5ef903e5e785
But well your Syphilis Brain has obviously
shut down its operation, silly Micro Penis.
Bye
Mild Shock schrieb:
Hi,
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
secure sand box, for executing JavaScript in the main
and in isolated siloed workers. Also you seem not to see
that I post CHF price tags and CHF product listings.
So what makes you think Switzerland == Germany,
when you rant like below. Even a dog knows that a
non-EU country cannot be the same as EU country:
you stupid half german, the comparisons along
memories arrays gpu card located, are taking
place parallel without cpu intervention, you
fucking illiterate idiot. You soon will become a
quarter german hence 3/4 russian old days,
historically. It's coming
German energy crisis caused by rCylack of Russian gasrCO rCo Merz
https://www.rt.com/news/643254-germany-crisis-russian-gas/
Bye
P.S.: EU = European Union
Audie Balaban schrieb:
Mild Shock wrote:with a
Maybe you could post some subtantial critique moron? Instead of
gibberish all the time. What does a cellphone number have to do
security,REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
you lying bitch, your arse is burning, you said google chrome
and gave them your cellphone number, fucking idiot. And you put wrong >>> -a> ports numbers along the localhost: idiot
Mild Shock schrieb:
Hi,
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN"
http://localhost:1234/api/v0/models
Response format
{
-a-a "object": "list",
-a-a "data": [
-a-a-a-a {
-a-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a-a "object": "model",
-a-a-a-a-a-a "type": "vlm",
-a-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
How confused is tiny winy penis?
For the 100th time the budget here:
11.4 Giga Lips with a Budget Laptop
Ryzen AI 7 350 w/ Radeon 860M https://github.com/Jean-Luc-Picard-2021/gigabudget
is a laptop and not a smartphone. It
has no cellphone number. And w/ means
integrated GPU on the silicon chip,
and not a GPU connected to the mainboard
via some PCI bus. The model is a acer
swift go, I already posted this info:
Swift Go 16 AI SFG16-61-R21J Notebook
Mild Shock wrote:location
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
cretin, they already have your phone number, the IMEI, email adr,
and everything, your friends included. Idiot, I cant even believe it.
Hi,
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
secure sand box, for executing JavaScript in the main
and in isolated siloed workers. Also you seem not to see
that I post CHF price tags and CHF product listings.
So what makes you think Switzerland == Germany,
when you rant like below. Even a dog knows that a
non-EU country cannot be the same as EU country:
you stupid half german, the comparisons along
memories arrays gpu card located, are taking
place parallel without cpu intervention, you
fucking illiterate idiot. You soon will become a
quarter german hence 3/4 russian old days,
historically. It's coming
German energy crisis caused by rCylack of Russian gasrCO rCo Merz https://www.rt.com/news/643254-germany-crisis-russian-gas/
Bye
P.S.: EU = European Union
Audie Balaban schrieb:
Mild Shock wrote:
Maybe you could post some subtantial critique moron? Instead of
gibberish all the time. What does a cellphone number have to do with a
REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
you lying bitch, your arse is burning, you said google chrome security, and gave them your cellphone number, fucking idiot. And you put wrong ports numbers along the localhost: idiot
Mild Shock schrieb:
Hi,
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN"
http://localhost:1234/api/v0/models
Response format
{
-a-a "object": "list",
-a-a "data": [
-a-a-a-a {
-a-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a-a "object": "model",
-a-a-a-a-a-a "type": "vlm",
-a-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
Hi,
The IMEI (International Mobile Equipment Identity)
is a 15-digit, globally unique serial number that
identifies every mobile phone.
Most laptops do not have an IMEI number. It only
exists if the device has an integrated cellular
modem (WWAN/LTE/5G card) for SIM cards.
Also in Switzerland people have fiber optic earth
cables for internet, and not some 5G over the air.
I got like 10 GBit/s fiber here in my home.
There are 96 internet service providers that offer that speed:
https://www.comparis.ch/telecom/zuhause/angebote/list?requestobject={%22products%22%3A[1]%2C%22onlyOffersWithoutMinimumDuration%22%3Afalse%2C%22internetSpeedTypes%22%3A[1]%2C%22connectionTypes%22%3A[1]%2C%22tvOptions%22%3A[]%2C%22landlineOptions%22%3A[]%2C%22providers%22%3A[]%2C%22showDiscountedOnly%22%3Afalse%2C%22addressCheckInfo%22%3A{%22AvailableCableProviders%22%3Anull%2C%22AvailableVdslSpeed%22%3Anull%2C%22AvailableFiberSpeed%22%3Anull%2C%22AvailableInit7Fiber%22%3Afalse%2C%22AvailableAllFiberProvidersExceptInit7%22%3Afalse}%2C%22address%22%3A{%22Zip%22%3Anull%2C%22Street%22%3Anull%2C%22StreetNumber%22%3Anull}}
Angebote f|+r Internet, TV und Festnetz-Telefon, sowie Kombiangebote
Its handy to download LLMs which have GB sizes.
But this laptop has no SIM Card, even not a e-SIM:
How confused is tiny winy penis?
For the 100th time the budget here:
11.4 Giga Lips with a Budget Laptop
Ryzen AI 7 350 w/ Radeon 860M https://github.com/Jean-Luc-Picard-2021/gigabudget
is a laptop and not a smartphone. It
has no cellphone number. And w/ means
integrated GPU on the silicon chip,
and not a GPU connected to the mainboard
via some PCI bus. The model is a acer
swift go, I already posted this info:
Swift Go 16 AI SFG16-61-R21J Notebook
https://www.acer.com/ch-de/laptops/swift/swift-go-16-ai-amd/pdp/NX.JCREZ.007
Bye
Roque Bahtinov schrieb:
Mild Shock wrote:
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
cretin, they already have your phone number, the IMEI, email adr,location
and everything, your friends included. Idiot, I cant even believe it.
Mild Shock schrieb:
Hi,
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
secure sand box, for executing JavaScript in the main
and in isolated siloed workers. Also you seem not to see
that I post CHF price tags and CHF product listings.
So what makes you think Switzerland == Germany,
when you rant like below. Even a dog knows that a
non-EU country cannot be the same as EU country:
you stupid half german, the comparisons along
memories arrays gpu card located, are taking
place parallel without cpu intervention, you
fucking illiterate idiot. You soon will become a
quarter german hence 3/4 russian old days,
historically. It's coming
German energy crisis caused by rCylack of Russian gasrCO rCo Merz
https://www.rt.com/news/643254-germany-crisis-russian-gas/
Bye
P.S.: EU = European Union
Audie Balaban schrieb:
Mild Shock wrote:with a
Maybe you could post some subtantial critique moron? Instead of
gibberish all the time. What does a cellphone number have to do
security,REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
you lying bitch, your arse is burning, you said google chrome
and gave them your cellphone number, fucking idiot. And you put wrong
ports numbers along the localhost: idiot
Mild Shock schrieb:
Hi,
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN"
http://localhost:1234/api/v0/models
Response format
{
-a-a "object": "list",
-a-a "data": [
-a-a-a-a {
-a-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a-a "object": "model",
-a-a-a-a-a-a "type": "vlm",
-a-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
Hi,
I feel really pitty for Micro Penis:
- RTFM means study, ask your Ukrainian neighbours
-a how its done, they have brilliant brains
- Only Yandex browser, go to Ukraine, they
-a have Chrome, Safari, Edge, etc..
- No AMD APU (*), go to Ukraine, you might
-a lay hands on a AI Laptop there, etc..
Good Luck!
Bye
(*) Direct corporate sales and distribution of
AMD processors and APUs to Russia have been officially
suspended since early 2022 due to sweeping U.S. and
international export controls and sanctions. https://www.tomshardware.com/news/intel-amd-nvidia-tsmc-russia-stop-chip-sales-ukraine-sanction
Bye
Mild Shock schrieb:
Hi,
Micro Penis, you possible still think when I wrote
mobile grade GPU, thats it is a GPU only for
smallphones, right?
Maybe stop your hobby interpretation of everything,
and RTFM . And try to get a professional understanding .
I admit it might take you a few months to
Keep up with the Kardashians so to speek, i.e. the
reality show of GPUs, and not what your Syphilis Brain
halucinates. To get the basic understanding
that is needed to follow what I post.
Good Luck!
Bye
RTFM = Abbreviation for rCyRead The Fucking ManualrCO.
http://www.catb.org/esr/jargon/html/R/RTFM.html
Keeping Up with the Kardashians
https://en.wikipedia.org/wiki/Keeping_Up_with_the_Kardashians
Mild Shock schrieb:
Hi,
But I already wrote like a 100times that
you can use WebGPU outside of a browser.
The same applies for Workers.
You can use WebGPU and Workers from within
node.js. No chrome browser (from google)
needed, or safari (from apple) or edge (from
microsoft). If you don't need a User Experience
(UX), i.e. if its enough to go headless with
your use case, for example implementing a
REST endpoint, you can use node.js for both
WebGPU and Workers. I posted the WebGPU headless
article, like 100times already:
Like WebAssembly before it, WebGPU has "escaped" the browser
https://blog.4dpipeline.com/client-side-ai-is-here-how-webgpu-transforms-your-gpu-server-economics
I also recently blogged about how to
use multiple workers from a main via browser
AND node.js. It not extremly difficult:
Parallel -C-WAM: JavaScript Workers as CPU Backend
https://medium.com/2989/5ef903e5e785
But well your Syphilis Brain has obviously
shut down its operation, silly Micro Penis.
Bye
Mild Shock schrieb:
Hi,
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
secure sand box, for executing JavaScript in the main
and in isolated siloed workers. Also you seem not to see
that I post CHF price tags and CHF product listings.
So what makes you think Switzerland == Germany,
when you rant like below. Even a dog knows that a
non-EU country cannot be the same as EU country:
you stupid half german, the comparisons along
memories arrays gpu card located, are taking
place parallel without cpu intervention, you
fucking illiterate idiot. You soon will become a
quarter german hence 3/4 russian old days,
historically. It's coming
German energy crisis caused by rCylack of Russian gasrCO rCo Merz
https://www.rt.com/news/643254-germany-crisis-russian-gas/
Bye
P.S.: EU = European Union
Audie Balaban schrieb:
Mild Shock wrote:with a
Maybe you could post some subtantial critique moron? Instead of
gibberish all the time. What does a cellphone number have to do
security,REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
you lying bitch, your arse is burning, you said google chrome
and gave them your cellphone number, fucking idiot. And you putwrong
ports numbers along the localhost: idiot
Mild Shock schrieb:
Hi,
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN"
http://localhost:1234/api/v0/models
Response format
{
-a-a "object": "list",
-a-a "data": [
-a-a-a-a {
-a-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a-a "object": "model",
-a-a-a-a-a-a "type": "vlm",
-a-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
Hi,
The IMEI (International Mobile Equipment Identity)
is a 15-digit, globally unique serial number that
identifies every mobile phone.
Most laptops do not have an IMEI number. It only
exists if the device has an integrated cellular
modem (WWAN/LTE/5G card) for SIM cards.
Also in Switzerland people have fiber optic earth
cables for internet, and not some 5G over the air.
I got like 10 GBit/s fiber here in my home.
There are 96 internet service providers that offer that speed:
https://www.comparis.ch/telecom/zuhause/angebote/list?requestobject={%22products%22%3A[1]%2C%22onlyOffersWithoutMinimumDuration%22%3Afalse%2C%22internetSpeedTypes%22%3A[1]%2C%22connectionTypes%22%3A[1]%2C%22tvOptions%22%3A[]%2C%22landlineOptions%22%3A[]%2C%22providers%22%3A[]%2C%22showDiscountedOnly%22%3Afalse%2C%22addressCheckInfo%22%3A{%22AvailableCableProviders%22%3Anull%2C%22AvailableVdslSpeed%22%3Anull%2C%22AvailableFiberSpeed%22%3Anull%2C%22AvailableInit7Fiber%22%3Afalse%2C%22AvailableAllFiberProvidersExceptInit7%22%3Afalse}%2C%22address%22%3A{%22Zip%22%3Anull%2C%22Street%22%3Anull%2C%22StreetNumber%22%3Anull}}
Angebote f|+r Internet, TV und Festnetz-Telefon, sowie Kombiangebote
Its handy to download LLMs which have GB sizes.
But this laptop has no SIM Card, even not a e-SIM:
How confused is tiny winy penis?
For the 100th time the budget here:
11.4 Giga Lips with a Budget Laptop
Ryzen AI 7 350 w/ Radeon 860M https://github.com/Jean-Luc-Picard-2021/gigabudget
is a laptop and not a smartphone. It
has no cellphone number. And w/ means
integrated GPU on the silicon chip,
and not a GPU connected to the mainboard
via some PCI bus. The model is a acer
swift go, I already posted this info:
Swift Go 16 AI SFG16-61-R21J Notebook
https://www.acer.com/ch-de/laptops/swift/swift-go-16-ai-amd/pdp/NX.JCREZ.007
Bye
Roque Bahtinov schrieb:
Mild Shock wrote:
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
cretin, they already have your phone number, the IMEI, email adr,location
and everything, your friends included. Idiot, I cant even believe it.
Mild Shock schrieb:
Hi,
Moron, from St. Petersburg, with only 5G internet.
You confuse browser security model with Google account.
No cellphone number involved in a brower JavaScript
secure sand box, for executing JavaScript in the main
and in isolated siloed workers. Also you seem not to see
that I post CHF price tags and CHF product listings.
So what makes you think Switzerland == Germany,
when you rant like below. Even a dog knows that a
non-EU country cannot be the same as EU country:
you stupid half german, the comparisons along
memories arrays gpu card located, are taking
place parallel without cpu intervention, you
fucking illiterate idiot. You soon will become a
quarter german hence 3/4 russian old days,
historically. It's coming
German energy crisis caused by rCylack of Russian gasrCO rCo Merz
https://www.rt.com/news/643254-germany-crisis-russian-gas/
Bye
P.S.: EU = European Union
Audie Balaban schrieb:
Mild Shock wrote:with a
Maybe you could post some subtantial critique moron? Instead of
gibberish all the time. What does a cellphone number have to do
security,REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
you lying bitch, your arse is burning, you said google chrome
and gave them your cellphone number, fucking idiot. And you put wrong
ports numbers along the localhost: idiot
Mild Shock schrieb:
Hi,
Maybe you could post some subtantial critique moron?
Instead of gibberish all the time. What does a cellphone
number have to do with a REST endpoint? Nothing!
Its all locally and my laptop has no cellphone number:
Start the REST API server
To start the server, run the following command:
lms server start
Endpoints
GET /api/v0/models
List all loaded and downloaded models
Example request
curl -H "Authorization: Bearer $LM_API_TOKEN"
http://localhost:1234/api/v0/models
Response format
{
-a-a "object": "list",
-a-a "data": [
-a-a-a-a {
-a-a-a-a-a-a "id": "qwen2-vl-7b-instruct",
-a-a-a-a-a-a "object": "model",
-a-a-a-a-a-a "type": "vlm",
-a-a-a-a-a-a "publisher": "mlx-community",
-a-a-a-a-a-a "arch": "qwen2_vl"
Etc...
https://lmstudio.ai/docs/developer/rest/endpoints
Bye
BTW: LM Studio recently introduced LM Link,
which provides some VPN. It can be used to
create clients or servers that run models.
It is end-to-end encrypted, and built on top
of custom Tailscale mesh VPNs. This is for
the paranoid, that want to acccess a
LLM from one end of the globe, that sits
on the other end of the globe, and have
no evesdroper or whatever on the
information that is exchanged.
I have nothing to hide,
you can find me in search.ch
Mild Shock wrote:
Since I don't belong to some post CCCP, cigaret smuggling cartell, I
have nothing to hide. If you are clever you find
hence you are admitting you are a fucking inbreed half german idiot from birth. Thanks making it clearer.
Hi,
I have nothing to hide,
you can find me in search.ch
Since when belongs .ch to German? The .ch
is the country code top-level domain (ccTLD)
for Switzerland in the Domain Name System
of the Internet.
If you want find proof of Germany, you
would need to see .de. That I use a German
usenet provider doesn't mean I am German.
Everybody can use solani.org.
Switzerland, SWITCH
https://en.wikipedia.org/wiki/.ch
Germany, DENIC
https://en.wikipedia.org/wiki/.de
nicht-kommerziellen Usenet-News-Server
https://solani.org/
Whats wrong with you?
Bye
Keiv Babenchikov schrieb:
Mild Shock wrote:
Since I don't belong to some post CCCP, cigaret smuggling cartell, I
have nothing to hide. If you are clever you find
hence you are admitting you are a fucking inbreed half german idiot from birth. Thanks making it clearer.
Hi,[...]
Because of MIMD you have to reassess algorithms.
A spin loop which could really hurt non-MIMD
GPUs, might less hurt a MIMD GPU.
On 7/21/2026 12:02 AM, Mild Shock wrote:
Hi,[...]
Because of MIMD you have to reassess algorithms.
A spin loop which could really hurt non-MIMD
GPUs, might less hurt a MIMD GPU.
Huh? atomic fetch-add has no looping. compare LOCK CMPXCHG with LOCK
XADD on x86.
Mild Shock wrote:
Hi,
CAS and XADD have no looping, they
are atomic operations, that take some
time but basically have some outcome
My father says C++ reminds him of an abortion.
Hi,
CAS and XADD have no looping, they
are atomic operations, that take some
time but basically have some outcome
with some ACID property and a result
value. What loops is the ADT, the Abstract
Data Type that you implement. Respectively
the client that uses the Abstract Data Type.
In your case you added the loop inside the
Abstract Data Type or lower level aggregate
code of a higher level operation:
Chris M. Thomasson wrote:
void producer(double state) {
-a-a-a uint32_t ver = XADD(&head, 1);
-a-a-a cell& c = cells[ver & (N - 1)];
-a-a-a while (LOAD(&c.ver) != ver) backoff(); /** Looping **/
-a-a-a c.state = state;
-a-a-a STORE(&c.ver, ver + 1);
}
https://groups.google.com/g/lock-free/c/acjQ3-89abE/m/a6-Di0GZsyEJ
In my case I added the loop during the client
usage of the ADT:
From: Mild Shock <janburse@fastmail.fm>
Subject: Source of the benchmark for DmitryVyukov
Date: Tue, 21 Jul 2026 01:44:21 +0200
-a-a-a private static void producer(Queue q) {
-a-a-a-a-a-a-a for (int i = 0; i < WORK; i++) {
-a-a-a-a-a-a-a-a-a-a-a Integer val = Integer.valueOf(i);
-a-a-a-a-a-a-a-a-a-a-a while (!enqueue(q, val)) ; /** Looping **/
-a-a-a-a-a-a-a }
-a-a-a }
Do you see the two loops, in your C code
and in my Java code? They are marked with a
comment /** Looping **/ .
You see them, don't you? But I don't know
exactly what backoff() does. Sometimes loops
are spurious yield loops, required because
an ADT cannot gurantee that every yield
implies a certain condition. This is for
example already found in the intrinsinc
monitor of Java, the wait(). You might consult
Doug Lea about the matter and how idiomatic
Java code looks like dealing with
spurious yields.
Bye
Chris M. Thomasson schrieb:
On 7/21/2026 12:02 AM, Mild Shock wrote:
Hi,[...]
Because of MIMD you have to reassess algorithms.
A spin loop which could really hurt non-MIMD
GPUs, might less hurt a MIMD GPU.
Huh? atomic fetch-add has no looping. compare LOCK CMPXCHG with LOCK
XADD on x86.
Hi,
CAS and XADD have no looping, they
are atomic operations, that take some
time but basically have some outcome
with some ACID property and a result
value. What loops is the ADT, the Abstract
Data Type that you implement. Respectively
the client that uses the Abstract Data Type.
In your case you added the loop inside the
Abstract Data Type or lower level aggregate
code of a higher level operation:
Chris M. Thomasson wrote:
void producer(double state) {
-a-a-a uint32_t ver = XADD(&head, 1);
-a-a-a cell& c = cells[ver & (N - 1)];
-a-a-a while (LOAD(&c.ver) != ver) backoff(); /** Looping **/
-a-a-a c.state = state;
-a-a-a STORE(&c.ver, ver + 1);
}
https://groups.google.com/g/lock-free/c/acjQ3-89abE/m/a6-Di0GZsyEJ
I think it must be possible to combine e.g. CAS-based consumers and
XADD-based producers. This can be beneficial if you expect that
producers usually do not blocks (so you care more about fast-path
performance rather than blocking behavior).
AFAICT, my tweak should be 100% compatible with the
existing algorithm as-is. IMVHO, it could be a fairly
beneficial addition to the existing API. It opens
up a new way to think about handling contention wrt
using the queue as a whole.
In my case I added the loop during the client
usage of the ADT:
From: Mild Shock <janburse@fastmail.fm>
Subject: Source of the benchmark for DmitryVyukov
Date: Tue, 21 Jul 2026 01:44:21 +0200
-a-a-a private static void producer(Queue q) {
-a-a-a-a-a-a-a for (int i = 0; i < WORK; i++) {
-a-a-a-a-a-a-a-a-a-a-a Integer val = Integer.valueOf(i);
-a-a-a-a-a-a-a-a-a-a-a while (!enqueue(q, val)) ; /** Looping **/
-a-a-a-a-a-a-a }
-a-a-a }
Do you see the two loops, in your C code
and in my Java code? They are marked with a
comment /** Looping **/ .
You see them, don't you? But I don't know
exactly what backoff() does. Sometimes loops
are spurious yield loops, required because
an ADT cannot gurantee that every yield
implies a certain condition. This is for
example already found in the intrinsinc
monitor of Java, the wait(). You might consult
Doug Lea about the matter and how idiomatic
Java code looks like dealing with
spurious yields.
Bye
Chris M. Thomasson schrieb:
On 7/21/2026 12:02 AM, Mild Shock wrote:
Hi,[...]
Because of MIMD you have to reassess algorithms.
A spin loop which could really hurt non-MIMD
GPUs, might less hurt a MIMD GPU.
Huh? atomic fetch-add has no looping. compare LOCK CMPXCHG with LOCK
XADD on x86.
On 7/22/2026 3:09 PM, Mild Shock wrote:
Hi,
CAS and XADD have no looping, they
are atomic operations, that take some
time but basically have some outcome
with some ACID property and a result
value. What loops is the ADT, the Abstract
Data Type that you implement. Respectively
the client that uses the Abstract Data Type.
In your case you added the loop inside the
Abstract Data Type or lower level aggregate
code of a higher level operation:
Chris M. Thomasson wrote:
void producer(double state) {
-a-a-a-a uint32_t ver = XADD(&head, 1);
-a-a-a-a cell& c = cells[ver & (N - 1)];
-a-a-a-a while (LOAD(&c.ver) != ver) backoff(); /** Looping **/
-a-a-a-a c.state = state;
-a-a-a-a STORE(&c.ver, ver + 1);
}
https://groups.google.com/g/lock-free/c/acjQ3-89abE/m/a6-Di0GZsyEJ
This was meant to be used in either a backoff, or ideally a futex, or
return the bakery ticket to the user in the need to wait case. Was never meant to be used in a GPU. Dmitry CAS version can be used, but its still going to spin on any failed CAS. But, they are very different ways to do
the same thing and they can be mixed and matched. But on the GPU, try to avoid CAS or anything that has to wait.
My XADD version beats Dmitry's in some work loads, but the fact it that
they can be mixed and matched.
_____________
[from me]
I think it must be possible to combine e.g. CAS-based consumers and
XADD-based producers. This can be beneficial if you expect that
producers usually do not blocks (so you care more about fast-path
performance rather than blocking behavior).
AFAICT, my tweak should be 100% compatible with the
existing algorithm as-is. IMVHO, it could be a fairly
beneficial addition to the existing API. It opens
up a new way to think about handling contention wrt
using the queue as a whole.
[from my friend]
Good point.
I guess single XADD for MPMC if far superior than most algorithms out
there, and is basically the best one can get for a centralized queue.
I see people still implement MS node-based queue with PDR, and that's
2 CAS loops + indirections + memory allocations + PDR acquire/release overheads.... veeeeeery slow :)
_______________
From this post:
https://groups.google.com/g/lock-free/c/acjQ3-89abE/m/q9j14HPbO3sJ
You need to be careful wrt just using a multi-threaded lock-free algo in
a compute shader. Some are just not made for it.
Why do you even need a mpmc queue in your compute shader anyway?
In my case I added the loop during the client
usage of the ADT:
From: Mild Shock <janburse@fastmail.fm>
Subject: Source of the benchmark for DmitryVyukov
Date: Tue, 21 Jul 2026 01:44:21 +0200
-a-a-a-a private static void producer(Queue q) {
-a-a-a-a-a-a-a-a for (int i = 0; i < WORK; i++) {
-a-a-a-a-a-a-a-a-a-a-a-a Integer val = Integer.valueOf(i);
-a-a-a-a-a-a-a-a-a-a-a-a while (!enqueue(q, val)) ; /** Looping **/
-a-a-a-a-a-a-a-a }
-a-a-a-a }
Do you see the two loops, in your C code
and in my Java code? They are marked with a
comment /** Looping **/ .
You see them, don't you? But I don't know
exactly what backoff() does. Sometimes loops
are spurious yield loops, required because[...]
Hi,
Because I use WebGPU and not WebGL. And
because WebGPU can adresss modern GPU
developed with the NVIDIA Volta evolution,
which happened in 2017. Namley that compute
shaders are not any more subject to the
realization restriction of lock step
execution, but have independent thread state.
And because there is independent thread state
there is also independent time spent for a
a work item by each logical thread, if the
submitted logical thread uses a lot of branching
logic or even loops. But the use of branching
and loops is encouraged in independent thread
state programming of compute shaders. The variables
that can drive such logic are the scalar variables:
Tour of WGSL - Control Flow https://google.github.io/tour-of-wgsl/control-flow/
Then not to waste GPU compute time, by logical
threads doing nothing. You will need to
introduce some load balancing among multiple
logical threads. And MPMC queues are one way to
readize load balancing. Compute shaders with
producer and consumer entry points are proposed
as fundamental architecture by Thunder Kittens:
ThunderKittens: Simple, Fast, and Adorable AI Kernels https://arxiv.org/abs/2410.20399
They are used by this SpaceX acquisition:
Composer 2 Technical Report
https://arxiv.org/abs/2603.24477
Thunder Kittens uses Hardware support, i.e. tma_expect().
Bye
Chris M. Thomasson schrieb:
never meant to be used in a GPU.
Dmitry CAS version can be used, but
Why do you even need a mpmc queue
in your compute shader anyway?
I don't think he knows exactly what he is doing...
Why does he need a lock/wait-free queue in a
compute shader? What is he trying to do?
Hi,
You don't pay attention, right! I am little
bit disappointed that your attention span is
near zero. I already posted:
From: Mild Shock <janburse@fastmail.fm>
Subject: Why do you even need a mpmc queue? [Thunder Kittens]
Date: Thu, 23 Jul 2026 08:43:03 +0200
Hi,
Because I use WebGPU and not WebGL. And
because WebGPU can adresss modern GPU
developed with the NVIDIA Volta evolution,
which happened in 2017. Namley that compute
shaders are not any more subject to the
realization restriction of lock step
execution, but have independent thread state.
And because there is independent thread state
there is also independent time spent for a
a work item by each logical thread, if the
submitted logical thread uses a lot of branching
logic or even loops. But the use of branching
and loops is encouraged in independent thread
state programming of compute shaders. The variables
that can drive such logic are the scalar variables:
Tour of WGSL - Control Flow https://google.github.io/tour-of-wgsl/control-flow/
Then not to waste GPU compute time, by logical
threads doing nothing. You will need to
introduce some load balancing among multiple
logical threads. And MPMC queues are one way to
readize load balancing. Compute shaders with
producer and consumer entry points are proposed
as fundamental architecture by Thunder Kittens:
ThunderKittens: Simple, Fast, and Adorable AI Kernels https://arxiv.org/abs/2410.20399
They are used by this SpaceX acquisition:
Composer 2 Technical Report
https://arxiv.org/abs/2603.24477
Thunder Kittens uses Hardware support, i.e. tma_expect().
Bye
Chris M. Thomasson schrieb:
never meant to be used in a GPU.
Dmitry CAS version can be used, but
Why do you even need a mpmc queue
in your compute shader anyway?
Chris M. Thomasson schrieb:
I don't think he knows exactly what he is doing...
Why does he need a lock/wait-free queue in a
compute shader? What is he trying to do?
Hi,
You can also deduce that I need comms,
from pi in pi-WAM, since pi refers to pi-calculus.
There is also a nice paper, that I have already posted:
A pi-calculus Specification of Prolog https://scispace.com/pdf/a-pi-calculus-specification-of-prolog-3qf2pf04ud.pdf
You asked yourself why no atomic and
only comms? So its as simple as 1+1=2.
But usenet people are usually slow as fuck.
Take your time. You could spin loop to ingest
the topic, i.e. try again in 3-4 months, for
example reading some of the paper. Although
I know thats a totally unrealistic request, asking
a troll to do RTFM and study something. They
rather make themselves a total laughing stock,
play stupid games, win usenet prizes.
Bye
Mild Shock schrieb:
Hi,
You don't pay attention, right! I am little
bit disappointed that your attention span is
near zero. I already posted:
From: Mild Shock <janburse@fastmail.fm>
Subject: Why do you even need a mpmc queue? [Thunder Kittens]
Date: Thu, 23 Jul 2026 08:43:03 +0200
Hi,
Because I use WebGPU and not WebGL. And
because WebGPU can adresss modern GPU
developed with the NVIDIA Volta evolution,
which happened in 2017. Namley that compute
shaders are not any more subject to the
realization restriction of lock step
execution, but have independent thread state.
And because there is independent thread state
there is also independent time spent for a
a work item by each logical thread, if the
submitted logical thread uses a lot of branching
logic or even loops. But the use of branching
and loops is encouraged in independent thread
state programming of compute shaders. The variables
that can drive such logic are the scalar variables:
Tour of WGSL - Control Flow
https://google.github.io/tour-of-wgsl/control-flow/
Then not to waste GPU compute time, by logical
threads doing nothing. You will need to
introduce some load balancing among multiple
logical threads. And MPMC queues are one way to
readize load balancing. Compute shaders with
producer and consumer entry points are proposed
as fundamental architecture by Thunder Kittens:
ThunderKittens: Simple, Fast, and Adorable AI Kernels
https://arxiv.org/abs/2410.20399
They are used by this SpaceX acquisition:
Composer 2 Technical Report
https://arxiv.org/abs/2603.24477
Thunder Kittens uses Hardware support, i.e. tma_expect().
Bye
Chris M. Thomasson schrieb:
never meant to be used in a GPU.
Dmitry CAS version can be used, but
Why do you even need a mpmc queue
in your compute shader anyway?
Chris M. Thomasson schrieb:
I don't think he knows exactly what he is doing...
Why does he need a lock/wait-free queue in a
compute shader? What is he trying to do?
Show an outline of what you need you compute shader to do?
Hi,
You can also deduce that I need comms,
from pi in pi-WAM, since pi refers to pi-calculus.
There is also a nice paper, that I have already posted:
A pi-calculus Specification of Prolog https://scispace.com/pdf/a-pi-calculus-specification-of-prolog-3qf2pf04ud.pdf
You asked yourself why no atomic and
only comms? So its as simple as 1+1=2.
But usenet people are usually slow as fuck.
Take your time. You could spin loop to ingest
the topic, i.e. try again in 3-4 months, for
example reading some of the paper. Although
I know thats a totally unrealistic request, asking
a troll to do RTFM and study something. They
rather make themselves a total laughing stock,
play stupid games, win usenet prizes.
Bye
Mild Shock schrieb:
Hi,
You don't pay attention, right! I am little
bit disappointed that your attention span is
near zero. I already posted:
From: Mild Shock <janburse@fastmail.fm>
Subject: Why do you even need a mpmc queue? [Thunder Kittens]
Date: Thu, 23 Jul 2026 08:43:03 +0200
Hi,
Because I use WebGPU and not WebGL. And
because WebGPU can adresss modern GPU
developed with the NVIDIA Volta evolution,
which happened in 2017. Namley that compute
shaders are not any more subject to the
realization restriction of lock step
execution, but have independent thread state.
And because there is independent thread state
there is also independent time spent for a
a work item by each logical thread, if the
submitted logical thread uses a lot of branching
logic or even loops. But the use of branching
and loops is encouraged in independent thread
state programming of compute shaders. The variables
that can drive such logic are the scalar variables:
Tour of WGSL - Control Flow
https://google.github.io/tour-of-wgsl/control-flow/
Then not to waste GPU compute time, by logical
threads doing nothing. You will need to
introduce some load balancing among multiple
logical threads. And MPMC queues are one way to
readize load balancing. Compute shaders with
producer and consumer entry points are proposed
as fundamental architecture by Thunder Kittens:
ThunderKittens: Simple, Fast, and Adorable AI Kernels
https://arxiv.org/abs/2410.20399
They are used by this SpaceX acquisition:
Composer 2 Technical Report
https://arxiv.org/abs/2603.24477
Thunder Kittens uses Hardware support, i.e. tma_expect().
Bye
Chris M. Thomasson schrieb:
never meant to be used in a GPU.
Dmitry CAS version can be used, but
Why do you even need a mpmc queue
in your compute shader anyway?
Chris M. Thomasson schrieb:
I don't think he knows exactly what he is doing...
Why does he need a lock/wait-free queue in a
compute shader? What is he trying to do?
Hi,
You don't pay attention, right! I am little
bit disappointed that your attention span is
near zero. I already posted:
From: Mild Shock <janburse@fastmail.fm>
Subject: Why do you even need a mpmc queue? [Thunder Kittens]
Date: Thu, 23 Jul 2026 08:43:03 +0200
Hi,
Because I use WebGPU and not WebGL. And
because WebGPU can adresss modern GPU
developed with the NVIDIA Volta evolution,
which happened in 2017. Namley that compute
shaders are not any more subject to the
realization restriction of lock step
execution, but have independent thread state.
And because there is independent thread state
there is also independent time spent for a
a work item by each logical thread, if the
submitted logical thread uses a lot of branching
logic or even loops. But the use of branching
and loops is encouraged in independent thread
state programming of compute shaders. The variables
that can drive such logic are the scalar variables:
Tour of WGSL - Control Flow https://google.github.io/tour-of-wgsl/control-flow/
Then not to waste GPU compute time, by logical
threads doing nothing. You will need to
introduce some load balancing among multiple
logical threads. And MPMC queues are one way to
readize load balancing. Compute shaders with
producer and consumer entry points are proposed
as fundamental architecture by Thunder Kittens:
ThunderKittens: Simple, Fast, and Adorable AI Kernels https://arxiv.org/abs/2410.20399
They are used by this SpaceX acquisition:
Composer 2 Technical Report
https://arxiv.org/abs/2603.24477
Thunder Kittens uses Hardware support, i.e. tma_expect().
Bye
Chris M. Thomasson schrieb:
never meant to be used in a GPU.
Dmitry CAS version can be used, but
Why do you even need a mpmc queue
in your compute shader anyway?
Chris M. Thomasson schrieb:
I don't think he knows exactly what he is doing...
Why does he need a lock/wait-free queue in a
compute shader? What is he trying to do?
| Sysop: | Amessyroom |
|---|---|
| Location: | Fayetteville, NC |
| Users: | 74 |
| Nodes: | 6 (0 / 6) |
| Uptime: | 46:33:46 |
| Calls: | 1,100 |
| Files: | 1,339 |
| Messages: | 275,493 |