Sujet : Re: Is Intel exceptionally unsuccessful as an architecture designer?
De : monnier (at) *nospam* iro.umontreal.ca (Stefan Monnier)
Groupes : comp.archDate : 23. Sep 2024, 21:30:49
Autres entêtes
Organisation : A noiseless patient Spider
Message-ID : <jwvr09aktrs.fsf-monnier+comp.arch@gnu.org>
References : 1 2 3 4 5 6 7 8 9 10 11 12 13
User-Agent : Gnus/5.13 (Gnus v5.13)
Chris M. Thomasson [2024-09-20 14:54:36] wrote:
I had this crazy idea of putting cpus right on the ram. So, if you add
more memory to your system you automatically get more cpu's... Think
NUMA for a moment... ;^)
To which Lawrence D'Oliveiro preemptively replied:
Yes, but that’s a lot more expensive.
🙂
More seriously: the idea of putting the CPU closer to the RAM and
bundling them has been with us since last century.
Until recently it was limited to situations like embedded systems,
because the lack of flexibility was a major downer. But it's now used
in many more circumstances, typically by stacking N dies of RAM on top
of a die of CPU, because with current RAM sizes and CPUs it can give a
respectable amount of RAM, with the advantage of a comfortable
memory bandwidth.
I'm not following GPUs very closely, but I'd be surprised if there
aren't any GPUs which have RAM in the same chip (via stacked dies).
But maybe these are still "lower tier" GPUs, which don't pay much
attention to working efficiently when N such chips are put into a single
system (IOW you may still be unable to just "add more RAM(w/GPU)").
Stefan