[Beowulf] Register article on Epyc (Brian Dobbins)
gus at ldeo.columbia.edu
Thu Jun 22 15:36:58 PDT 2017
On 06/22/2017 05:04 PM, John Hearns wrote:
> David, you recall correctly. I recall working with Clustervision. We
> installed the first 64 bit x86 cluster in the UK,
> at the chemistry department in Manchester University. AMD CPUs, 1U
> pizza boxes.
We had a Sun Linux cluster with Opteron 244 (?) (SunFire v20 servers?)
here at about that time (2004).
Fancy back then, with "service processors" (kind of IPMI), etc.
They're not even on eBay anymore.
> For the life of me I Cannot recall the manufacturer... but it was a
> white box.
> Remember that SuSE hired the guy who ported to the x86-64
> architecture. I remember him giving LUG talks...
> (arghhh.. name escapes me too...)
Oh, don't worry about the lapse.
Memory bandwidth is an AMD thing only.
We have a 2013 AMD Abu Dhabi/Piledriver cluster, and
the two-core-shared floating point unit makes it
perform worse than comparable Intel processors
of that time (Sandy Bridge) ... but it cost significantly less,
and still supports our computational needs well.
Maybe with these new Epyc we can go back to buying
non-bleeding-edge processors from the underdog at more affordable prices,
and still run memory-greedy weather/climate, CFD, computational chemistry
codes with a decent performance.
> On 22 June 2017 at 21:27, mathog <mathog at caltech.edu
> <mailto:mathog at caltech.edu>> wrote:
> On Thu, Jun 22, 2017 at 10:04:34 -0600 Brian Dobbins wrote
> On Wed, Jun 21, 2017 at 6:29 PM, Christopher Samuel
> <samuel at unimelb.edu.au <mailto:samuel at unimelb.edu.au>>
> I thought it interesting that the only performance info in
> that article
> for Epyc were SpecINT and (the only mention for SpecFP was
> for Radeon).
> As did I, but a little digging shows a STREAM benchmark (on
> AMD's page)
> showing +25% performance of a single-socket Epyc vs. a
> dual-socket E5-2690
> v4 Broadwell system, and roughly 60% better specfp_rate2006
> numbers when
> comparing socket-to-socket. When I read stuff like this, I
> feel a
> little bit like Charlie Brown going to kick the football, and
> *hoping* it's
> not going to get whisked away by Lucy...
> If by Lucy you mean Intel, then that suspicion may have some merit.
> Recall that when the Opterons first came out the major
> manufacturers did not ship any systems with it for what, a year,
> maybe longer? I vaguely recall SuperMicro going in quickly and
> Dell, HP, and IBM whistling in a corner. Something about
> contractual obligations to Intel, or a desire not to piss off Intel.
> Let's see, HP shipped its first system "in the first half" of 2004.
> while the first Opterons shipped in, um, April 2003. So yes,
> about a year.
> When the multiple core Opterons came out once again the big
> manufacturers were slow to ship them, although in some cases it
> was apparently due to supply issues:
> David Mathog
> mathog at caltech.edu <mailto:mathog at caltech.edu>
> Manager, Sequence Analysis Facility, Biology Division, Caltech
> Beowulf mailing list, Beowulf at beowulf.org
> <mailto:Beowulf at beowulf.org> sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
More information about the Beowulf