We use the per core SI2K summed over cores - and each generation of
systems contributes a scaled number appropriate to its performance. The
numbers published here agree with what I calculate using that method
so...
As a member of the HEPiX group...
The HEPiX group have concluded that a benchmark based on selected
measures available within the SPEC2006 suite is the way to go, obtained
in a specific way (compiler flags, simultaneous runs one per core etc).
The chosen benchmark has the advantage that it isn't one of the standard
reportable / published benchmarks (neither INT or FP) so vendors have to
measure a system and cannot cheat. It also means we have to benchmark
the kit too, though a site licence for the SPEC suite is not expensive
and the benchmark is trivial to run.
Martin.
--
Martin Bly
RAL Tier1 Fabric Team
> -----Original Message-----
> From: Testbed Support for GridPP member institutes
> [mailto:[log in to unmask]] On Behalf Of Burke, S (Stephen)
> Sent: 31 October 2008 12:25
> To: [log in to unmask]
> Subject: Specint
>
> Hi,
>
> There's currently a debate going on about the meaning of the specint
> value published in the information system, and I'd like to get some
> feedback on what people are actually doing. The question is
> whether, for
> multi-core CPUs, the specint rating applies to a single core or to a
> whole CPU (aside from technical details about how you measure
> it). What
> are people actually doing?
>
> Stephen
> --
> Scanned by iCritical.
>
--
Scanned by iCritical.
|