Re: [AMBER] performance of pmed.cuda.MPI

From: Tru Huynh <tru.pasteur.fr>
Date: Sun, 23 Sep 2012 22:18:57 +0200

Thanks you all for this discussion :)

I was looking at this kind of gear:
http://www.supermicro.nl/products/system/4u/7047/sys-7047gr-tprf.cfm

4x PCI-E 3.0 16x buses so enough bandwith for 4x GTX680.
The power supply in the 1600 W range should be enough

The 300W GTX690 are probably too much drain (both thermal and power)
and with the shared PCI-E bandwith issue, that's not really good! :D

I might go from lower CPU specs, 2 sockets at 4 cores instead of
the high end 6/8 cores.
Of course they will cost more than your spec'ed GPU dedicated boxes
http://www.rosswalker.co.uk/current_amber_gpu_spec.htm

I am trading space (4U rack/4 gpus) and remote management (ipmi/kvm).
Since this is for one proof of concept box before choosing
K10/K20 GTX680 or plain M2090 for a few more boxes.

I really enjoyed your [AMBER] Tesla or GTX? thread reply!

Best regards,

Tru
-- 
Dr Tru Huynh          | http://www.pasteur.fr/recherche/unites/Binfs/
mailto:tru.pasteur.fr | tel/fax +33 1 45 68 87 37/19
Institut Pasteur, 25-28 rue du Docteur Roux, 75724 Paris CEDEX 15 France  
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Sep 23 2012 - 13:30:02 PDT
Custom Search