Re: run tests with CUDA

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Mon May 02 2011 - 16:55:18 CDT

On Mon, May 2, 2011 at 5:25 PM, Massimiliano Porrini <M.Porrini_at_ed.ac.uk> wrote:
> Dear all,
>
> I was testing NAMD-CUDA (precisely NAMD_2.8b1_Linux-x86_64-CUDA)
> on my GPU workstation which has:
>
> - two Tesla C2070 boards
>
> - one GeForce GTX 470 card
>
> - Intel Xeon E5620 Processor, 2.40GHz (Quad-Core/8 thread SMT)
>
> The system under investigation is an 11 residues long peptide immersed
> in a TIP3P
> water molecules box (truncated octahedron), with a total number of atoms equal
> to 14038. The simulation conf file is attached.
>
> I would like to know if the following computing times are OK,
> or there are some improving settings (hardware or software)
> that I am missing in my machine:
>
> - CPU run with 16 processes: 14m 29s
> - GeForce GTX 470:                20m 33s
> - 1 Tesla:                                  21m 02s
> - 2 Teslas:                                13m 02s
> - 2 Teslas + GTX 470:             10m 13s
>
> I am asking this mainly because I would expect that one Tesla C2070
> was anyway faster than one GeForce GTX 470 and 2 Teslas faster than
> the CPU, am I wrong?

yes and no. GPUs are "threading monsters" and thus need
to execute as many threads as possible to be efficient.

you have far too few atoms to get that much of a speedup.
try the same test again with a larger system.

the second issue is that you have to move a lot of data in and
out of the GPU and those GPUs may be fighting over PCI-e
bandwidth. typical mainboards only have enough aggregate
PCI-e bandwidth for two GPUs, even if they have more
"full 16x PCI-e-v2.0" slots.

you can also try to run two NAMD instances per GPU,
but it may not be helping a lot in your case.

cheers,
   axel.

>
> Many thanks in advance.
>
> Best regards,
> MP
>
>
>
> --
> Dr Massimiliano Porrini
> P. E. Barran Research Group
> Institute for Condensed Matter and Complex Systems
> School of Physics & Astronomy
> The University of Edinburgh
> James Clerk Maxwell Building
> The King's Buildings
> Mayfield Road
> Edinburgh EH9 3JZ
>
> Tel +44-(0)131-650-5229
>
> E-mails : M.Porrini_at_ed.ac.uk
>              mozz76_at_gmail.com
>              maxp_at_iesl.forth.gr
>

-- 
Dr. Axel Kohlmeyer
akohlmey_at_gmail.com  http://goo.gl/1wk0
Institute for Computational Molecular Science
Temple University, Philadelphia PA, USA.

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:20:11 CST