By: juanrga (nospam.delete@this.juanrga.com), January 29, 2017 6:13 am
Room: Moderated Discussions
wumpus (lost.delete@this.in-a.cave.net) on January 22, 2017 11:31 am wrote:
> juanrga (noemail.delete@this.juanrga.com) on January 22, 2017 2:59 am wrote:
> > Dumb belief? ROFL
>
> I'd be really curious how in the world you would expect to get all the memory and switches to run at low
> enough power to justify the hit in performance. What stupendously powerful ARM floating point monster
> is out there to build a supercomputer? It is the same reason ARM can't supply power/wattage in the server
> room (except now with floating point issues), even if the CPU itself might use less power, there are other
> bits in the system that consume power at intel levels while limited to ARM performance.
What is really curious is how data has been published again and again is ignored.
> It is hardly surprising that while nVidia is busy designing ARMs, they never bother designing
> their supercomputer parts to interface with anything other than Intel and IBM CPUs.
No sure if you are claiming that commercial ARM+Nvidia HPC clusters don't exist or if you claim that Nvidia never shared slides like this

> juanrga (noemail.delete@this.juanrga.com) on January 22, 2017 2:59 am wrote:
> > Dumb belief? ROFL
>
> I'd be really curious how in the world you would expect to get all the memory and switches to run at low
> enough power to justify the hit in performance. What stupendously powerful ARM floating point monster
> is out there to build a supercomputer? It is the same reason ARM can't supply power/wattage in the server
> room (except now with floating point issues), even if the CPU itself might use less power, there are other
> bits in the system that consume power at intel levels while limited to ARM performance.
What is really curious is how data has been published again and again is ignored.
> It is hardly surprising that while nVidia is busy designing ARMs, they never bother designing
> their supercomputer parts to interface with anything other than Intel and IBM CPUs.
No sure if you are claiming that commercial ARM+Nvidia HPC clusters don't exist or if you claim that Nvidia never shared slides like this
