ALL NPPS Members

[Johannes Elmsheuser <johannes.elmsheuser AT cern.ch>]

Hi Brett,

On 16.07.25 13:52, Brett Viren wrote:
> Really interesting.
>
> The plot shows a clear motivation to use ARM for reasons of electricity.

Things are always in flux - the very latest AMD EPYC processors have 
caught up in that metric.

> Is there a similar plot that shows HS23 score / "initial cost"?  My
> guess (and Gemini seems to agree) is that ARM systems are also cheaper
> (per unit compute).

No, we do not have such a plot.

AWS is running a significant part of their compute on their in house 
Graviton ARM processors (most probably because of cost).

> The higher ARM core count at first might be a constraint but the ARM
> systems seem to come with even higher RAM per core than I'm used to
> seeing on existing Intel "grid" slots for DUNE.

The ARM processors are btw. not hyperthreaded so there is an even better 
throughput compared to AMD/Intel - though the Nvidia grace chips are 
really expensive.
> Why are we not all using ARM!?

Cheers, Johannes
> -Brett.
>
>
> Johannes Elmsheuser (via phys-npps-members-l Mailing List)
> <phys-npps-members-l AT lists.bnl.gov> writes:
>
> > Hi all,
> >
> > there is now a short ATLAS experiment briefing article online about 'Extending
> > the ARM of ATLAS computing':
> > https://urldefense.com/v3/__https://atlas.cern/Updates/Briefing/ARM-for-ATLAS__;!!P4SdNyxKAPE!C28oeCyn7iiJtIKC5_P-LDnwmtP3JrMD_07R_denFoEklJTdrOblNTgQCBNQcUTYCvI1afSAv2L7BWXyFrgQwHtmlKwxU71TblliefmKgsqD$
> > Enjoy !
> > Cheers, Johannes