| Sysop: | Amessyroom |
|---|---|
| Location: | Fayetteville, NC |
| Users: | 42 |
| Nodes: | 6 (0 / 6) |
| Uptime: | 01:16:45 |
| Calls: | 220 |
| Calls today: | 1 |
| Files: | 824 |
| Messages: | 121,522 |
| Posted today: | 6 |
https://www.thehindu.com/sci-tech/technology/new-nvidia-ai-chips-face-issue-with-overheating-servers-report/article68880681.ece
Nvidia's new Blackwell AI chips, which have already faced
delays, have encountered problems with accompanying servers
that overheat, causing some customers to worry they will not
have enough time to get new data centres up and running,
the Information reported on Sunday.
The Blackwell graphics processing units overheat when
connected together in the customised server racks the
company has designed, the report said, citing sources
familiar with the issue.
. . .
Oh oh ... Nvidia kinda bet the bank on these
newest chips ......... and there are massive
downstream stock-market implications.
1000 watts TDP?
Seems just making more powerful chips by brute force has to end soon.
186282@ud0s4.net wrote:
https://www.thehindu.com/sci-tech/technology/new-nvidia-ai-chips-face-issue-with-overheating-servers-report/article68880681.ece
Nvidia's new Blackwell AI chips, which have already faced
delays, have encountered problems with accompanying servers
that overheat, causing some customers to worry they will not
have enough time to get new data centres up and running,
the Information reported on Sunday.
The Blackwell graphics processing units overheat when
connected together in the customised server racks the
company has designed, the report said, citing sources
familiar with the issue.
. . .
Oh oh ... Nvidia kinda bet the bank on these
newest chips ......... and there are massive
downstream stock-market implications.
1000 watts TDP?
Seems just making more powerful chips by brute force has to end soon.
On Thu, 21 Nov 2024 07:35:47 +0800, Woozy Song wrote:
1000 watts TDP?
Seems just making more powerful chips by brute force has to end soon.
Faster == hotter. I remember a seminar about 40 years ago when CMOS was coming into its own and touted as low power. The dirty little secret was
the power required to switch capacitive loads is a function of the voltage and frequency so the dynamic power dissipation exceeded that of 74LS parts when you drove them hard.
A lot has changed since then but I'm not waiting for a miracle.
On Thu, 21 Nov 2024 07:35:47 +0800
Woozy Song <suzyw0ng@outlook.com> wrote:
1000 watts TDP?
Seems just making more powerful chips by brute force has to end
soon.
We can't stop *now,* we've poured too much money into it...!
On Thu, 21 Nov 2024 07:35:47 +0800, Woozy Song wrote:
1000 watts TDP?
Seems just making more powerful chips by brute force has to end soon.
Faster == hotter. I remember a seminar about 40 years ago when CMOS was coming into its own and touted as low power. The dirty little secret was
the power required to switch capacitive loads is a function of the voltage and frequency so the dynamic power dissipation exceeded that of 74LS parts when you drove them hard.
A lot has changed since then but I'm not waiting for a miracle.
On 2024-11-21 05:45, rbowman wrote:
On Thu, 21 Nov 2024 07:35:47 +0800, Woozy Song wrote:
1000 watts TDP?
Seems just making more powerful chips by brute force has to end
soon.
Faster == hotter. I remember a seminar about 40 years ago when CMOS was
coming into its own and touted as low power. The dirty little secret
was the power required to switch capacitive loads is a function of the
voltage and frequency so the dynamic power dissipation exceeded that of
74LS parts when you drove them hard.
Still, they are fantastic for things like a battery powered clock. Or processors that stay a lot of the time waiting.
On Wed, 20 Nov 2024, John Ames wrote:
On Thu, 21 Nov 2024 07:35:47 +0800 Woozy Song <suzyw0ng@outlook.com>Anyone here shorting nvidia? Seems like a nice little bubble is building
wrote:
1000 watts TDP?
Seems just making more powerful chips by brute force has to end
soon.
We can't stop *now,* we've poured too much money into it...!
up there.
On Thu, 21 Nov 2024 09:52:27 +0100, D wrote:
On Wed, 20 Nov 2024, John Ames wrote:
On Thu, 21 Nov 2024 07:35:47 +0800 Woozy Song <suzyw0ng@outlook.com>Anyone here shorting nvidia? Seems like a nice little bubble is building
wrote:
1000 watts TDP?
Seems just making more powerful chips by brute force has to end
soon.
We can't stop *now,* we've poured too much money into it...!
up there.
I don't do stocks but that seems like a good bet. AMD and others are
coming from behind and may be able to supplant CUDA.
On Thu, 21 Nov 2024 12:04:12 +0100, Carlos E.R. wrote:
On 2024-11-21 05:45, rbowman wrote:
On Thu, 21 Nov 2024 07:35:47 +0800, Woozy Song wrote:
1000 watts TDP?
Seems just making more powerful chips by brute force has to end
soon.
Faster == hotter. I remember a seminar about 40 years ago when CMOS was
coming into its own and touted as low power. The dirty little secret
was the power required to switch capacitive loads is a function of the
voltage and frequency so the dynamic power dissipation exceeded that of
74LS parts when you drove them hard.
Still, they are fantastic for things like a battery powered clock. Or
processors that stay a lot of the time waiting.
Definitely. Like all design decisions they have their place. To really
date myself I sent my Osborne 1 back to the factory for the 100 column and DS/DD floppy upgrade. The new video would work for a while and then die. I isolated the problem chip with a can of cooler. It was CMOS and when I replaced it with the equivalent 74LS problem solved.
On 2024-11-21 20:59, rbowman wrote:
Definitely. Like all design decisions they have their place. To really
date myself I sent my Osborne 1 back to the factory for the 100 column
and
DS/DD floppy upgrade. The new video would work for a while and then
die. I
isolated the problem chip with a can of cooler. It was CMOS and when I
replaced it with the equivalent 74LS problem solved.
That's a curious one. :-)
Imagine my surprise at encountering 'commercial design' where it was sufficient if the thing worked between +10°C and +40°C and the customer
did the product testing, in terms of returned items.
On Wed, 27 Nov 2024 12:46:12 +0000, The Natural Philosopher wrote:
Imagine my surprise at encountering 'commercial design' where it was
sufficient if the thing worked between +10°C and +40°C and the customer
did the product testing, in terms of returned items.
My youthful idealism took a hit with my engineering statistics course.
Much time was devoted to how many widgets have to be tested to insure only
X% are crap, X being determined by the cost of replacing defective widgets versus the cost of doing it right.
On Wed, 27 Nov 2024 12:46:12 +0000, The Natural Philosopher wrote:
Imagine my surprise at encountering 'commercial design' where it was
sufficient if the thing worked between +10°C and +40°C and the customer
did the product testing, in terms of returned items.
My youthful idealism took a hit with my engineering statistics course.
Much time was devoted to how many widgets have to be tested to insure only
X% are crap, X being determined by the cost of replacing defective widgets versus the cost of doing it right.
On Thu, 21 Nov 2024, rbowman wrote:
On Thu, 21 Nov 2024 09:52:27 +0100, D wrote:
On Wed, 20 Nov 2024, John Ames wrote:
On Thu, 21 Nov 2024 07:35:47 +0800 Woozy Song <suzyw0ng@outlook.com>Anyone here shorting nvidia? Seems like a nice little bubble is building >>> up there.
wrote:
1000 watts TDP?
Seems just making more powerful chips by brute force has to end
soon.
We can't stop *now,* we've poured too much money into it...!
I don't do stocks but that seems like a good bet. AMD and others are
coming from behind and may be able to supplant CUDA.
I think once the LLM:s are exhausted and we no longer see significant
gains, might be a good time to start.
I think once the LLM:s are exhausted and we no longer see significant
gains, might be a good time to start.
LLMs can become very good - at least for many, human-replacement,
purposes. This is why Big Biz puts so much money into them, to
get rid of the annoying expensive humans.
Forgetting that disemployed humans can't BUY their stuff ...
(I suspect a "can't get there from here" point)
However I do think there's something beyond LLMs. Alas
I'm not 100% sure what that is. NN research continues,
but it doesn't yield as much as originally promised.
Might be even better ways of faking what NNs do without
trying to literally emulate the biological product -
at some point it's gonna be a kinda tight functional
equation. Actually real NNs - NOT just neurons/synapses
but lots of input from OTHER kinds of brain cells and
chemistry. Took a billion years of trial and error.
Proper 'consciousness' - don't know if it's the best
idea actually, won't be remotely as controllable as
optimists believe. However I suspect some kind of
'factal' infinite-regression/mirroring thing is involved.
It IS some kind of pattern/equation though, you can
smell it ... we see the paradigm not only in humans but
well down the tree. With varying degrees of sophistication
the "I AM" thing is widely seen.
TRUE "AI" ... we'd be building "aliens" - not at ALL
like we are.
It IS some kind of pattern/equation though, you can
smell it ... we see the paradigm not only in humans but
well down the tree. With varying degrees of sophistication
the "I AM" thing is widely seen.
On 03/12/2024 06:14, 186282@ud0s4.net wrote:
It IS some kind of pattern/equation though, you can
smell it ... we see the paradigm not only in humans but
well down the tree. With varying degrees of sophistication
the "I AM" thing is widely seen.
If you think about it, consciousness has to be separate from the thing
it is conscious of.
This leads to the 'problem of consciousness' in quantum physics.
On Tue, 3 Dec 2024, 186282@ud0s4.net wrote:
I think once the LLM:s are exhausted and we no longer see significant
gains, might be a good time to start.
LLMs can become very good - at least for many, human-replacement,
purposes. This is why Big Biz puts so much money into them, to
get rid of the annoying expensive humans.
So far I've found the AI:s I've encountered at companies to be very bad and annoying. I also read that more and more people support companies that provide
them with humans, over chat bots, and that the AI label is scaring
customers
away.
Forgetting that disemployed humans can't BUY their stuff ...
Like every technology shift, they will find other jobs, and our goods and services will become cheaper thanks to automation and AI, which will
benefit
everyone. There is no impending mass unemployment, regardless of how much politicians would like for that to be true, so they could scare people into accepting new "AI-taxes".
(I suspect a "can't get there from here" point)
However I do think there's something beyond LLMs. Alas
This is obvious. Of course, fast forward, 10-20 years, there will be
more break
throughs. I believe LLMs have hit their limits and that we won't see _dramatic_
gains as long as LLMs are the underlying technology. We will still see
gains,
but they are starting to taper off.
We will also see gains in pruning and making models smaller and more efficient,
while providing the same performance. This I also think is a given.
But LLMs will not lead us to consciousness. They lack will, volition and motivations. These things, for me, are important ingredients if you want to argue that you have created a consciousness.
I'm not 100% sure what that is. NN research continues,
but it doesn't yield as much as originally promised.
Might be even better ways of faking what NNs do without
trying to literally emulate the biological product -
at some point it's gonna be a kinda tight functional
equation. Actually real NNs - NOT just neurons/synapses
but lots of input from OTHER kinds of brain cells and
chemistry. Took a billion years of trial and error.
Proper 'consciousness' - don't know if it's the best
idea actually, won't be remotely as controllable as
optimists believe. However I suspect some kind of
'factal' infinite-regression/mirroring thing is involved.
It IS some kind of pattern/equation though, you can
smell it ... we see the paradigm not only in humans but
well down the tree. With varying degrees of sophistication
the "I AM" thing is widely seen.
TRUE "AI" ... we'd be building "aliens" - not at ALL
like we are.
Maybe. We don't even know what we are aiming at, so could be that we create something brand new and alien. Could also be that we run into some "laws of thought" that hit once the complexity of the system reaches a certain
level, and
we end up replicating ourselves.
I believe I will be alive when it happens and I'm looking forward to it! =)