From Newsgroup: sci.math
Hi,
Mostlikely the putin troll has still a
mental model of a GPU from the times
of Sputnik. But GPUs have evolved
from SIMD to MIMD. I posted that already
at least 3x times that a NVIDIA Volta
architecture is necessary for pi-WAM.
If I am not mistaken the evolution
happened around 10 years ago. It was
first announced on a roadmap in March
2013, although the first product was
not announced until May 2017. And its
amazingly what a stack overflow user writes:
"An important purpose of the feature is
to enable reliable use of programming models
such as producer-consumer within a warp"
https://stackoverflow.com/q/70987051
Thats exactly what I will explore,
in the context of pi-WAM. Maybe not
exactly as the stack overflow user
thinks it is done, but in spirit.
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.
--- Synchronet 3.22a-Linux NewsLink 1.2