[Beowulf] bring back 2012?

[Stu Midgley]

oh, indeed the top bin Xeon systems are fast and damn expensive.  Even when
we purchased these AMD systems they were a LOT cheaper than any Intel
system that could come close in specfp rate performance - let alone after 4
years of heavy use.

One issue is that the AMD systems are far more numa, thus requiring tighter
programming.

I was trying to justify tossing the AMD systems and hopefully replacing
them with a smaller number of broadwell based systems.  I guess I was
really surprised to see that a reasonable dual socket (what we would likely
purchase) broadwell system probably wasn't any faster.  Of course, per
socket, they are about 2x thus saving us a lot of space+power, but not the
~10x I was hoping to see.  It has been 4 years...


On Tue, Aug 16, 2016 at 11:49 PM, Elken, Tom <tom.elken at intel.com> wrote:

> You didn’t say on what basis you chose those two results, comparing a
> 4-socket, 64-core 2012 result to a 2-socket, 20-core 2016 result.
>
>
>
> Here’s a 4-socket 2016 result that looks like > 2x progress since the 2012
> result:
>
> https://www.spec.org/cpu2006/results/res2016q3/cpu2006-20160725-43005.html
>
> This _*would*_ be an expensive system.
>
>
>
> There are a lot of 2-socket results that are a ~50% improvement over the
> 4-socket 2012 base result you pointed to, e.g.:
>
> https://www.spec.org/cpu2006/results/res2016q3/cpu2006-20160705-42720.html
>
>
>
> But I’m glad you have thousands of Phi’s too.
>
>
>
> -Tom
>
>
>
> *From:* Beowulf [mailto:beowulf-bounces at beowulf.org] *On Behalf Of *Stu
> Midgley
> *Sent:* Tuesday, August 16, 2016 1:36 AM
> *To:* Beowulf List
> *Subject:* [Beowulf] bring back 2012?
>
>
>
> https://www.spec.org/cpu2006/results/res2016q2/cpu2006-20160308-39354.html
>
> https://www.spec.org/cpu2006/results/res2012q4/cpu2006-20121108-25077.html
>
>
>
> Its like no progress has been made.  So glad that we have thousands of
> Phi's...
>
>
>
> --
>
> Dr Stuart Midgley
> sdm900 at sdm900.com
>



-- 
Dr Stuart Midgley
sdm900 at sdm900.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20160817/bf289158/attachment.html>