[gmx-users] Benchmark performance question...

[Stéphane Teletchéa]

David van der Spoel a écrit :
> Stéphane Teletchéa wrote:
> 
>> Since my mail did not get much answer, i presume it was not clear 
>> enough...
>>
>> I'm benchmarking my system and i had these results (for villin as in 
>> gmxbench-3.0.tar.gz) :
>>                       NbCPU     Time (real)        Perf (ns/day)
>> single precision :     1          109.00           8.463
>> single precision :     2          102.00           8.471
>> double precision :     1          150.00           5.915
>> double precision :     2          166.00           5.205
>>
>> Is it possible to have only a small increase (single precision) on 1 
>> vs 2 processors (single precision) or do i need to multiply by nb of 
>> processors (two) the reported performance for 2-cpus ?
>>
>> Is there a file where this total performance is stored ?
> 
> It looks like you didn't run in parallel at all...

Ah, this is what i thought also...

> Is this a dual Xeon machine? Hyperthreading (on a P4) will make it slower.

Well my fault, i was sure i had checked that ht was not present in cat 
/proc/cpuinfo but it seems my eyes are more prone to error thant grep :
  cat /proc/cpuinfo |grep ht
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge 
mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid xtpr
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge 
mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe cid xtpr

Actually, it *is* a 2.6 Ghz PIV with HT ...

Results become clearer ...

Stef

-- 
Stéphane Téletchéa, PhD.                  http://www.steletch.org
Unité Mathématique Informatique et Génome http://migale.jouy.inra.fr/mig
INRA, Domaine de Vilvert                  Tél : (33) 134 652 121 / 3086
78352 Jouy-en-Josas cedex, France         Fax : (33) 134 652 901