[Beowulf] its going to be big
hearnsj at googlemail.com
Fri Oct 12 00:26:03 PDT 2018
Stu, congratulations on that project. There must have been an enormous
amount of effort behind the scenes there.
Wish I could be helping with it!
I am slightly intrigued by the emphasis on single precision FLOPS. Is that
because the codes run with single precision,
or is it that you find this a good metric of performance? I note no Top500
reference in the press release - you concentrate on the capability
of the service for your customer (*)
I guess I would extend that to comment that it is often said that Amazon,
Azure, Google etc. all have the raw compute power to place very highly in
the Top 500.
However it is probably not worth their time to (a) do the engineering to
get HPL running across datacentres and (b) set aside time on huge numbers
of servers which should be making money for them should.
Thinking out loud, I guess this is reflected in the traditional HPC Top500
- Top500 runs normally being doen in the commissioning phase as a good
workout for the system, and indeed again in traditional HPC the ratio of
achieved to theoretical peak is a good metric - too low an you know
something is wrong.
(*) I just did a sanity check. The Top500 site does say: . In particular,
the operation count for the algorithm must be 2/3 n^3 + O(n^2) double
precision floating point operations.
On Fri, 12 Oct 2018 at 07:57, Stu Midgley <sdm900 at gmail.com> wrote:
> Dr Stuart Midgley
> sdm900 at gmail.com
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Beowulf