[gmx-users] mdrun_gpu on NVidia Quadro 2000

[Mauricio Carrillo Tripp]

Hi,

I was wondering if the numbers I get are reasonable while running the DHFR
benchmark at:
http://www.gromacs.org/Documentation/Installation_Instructions/GPUs

As you can see, CPU performance is quite good, but GPU performance is
really bad
compared with results on that page for other cards (Tesla, GTX):

Performance [ns/day]:
*bench            CPU     GPU*
dhfr-impl-1nm    79.950  45.385
dhfr-impl-2nm    25.072  44.961
dhfr-impl-inf    17.597  25.737
dhfr-solv-PME    24.168   4.454
dhfr-solv-RF-1nm 33.339   8.594
dhfr-solv-RF-2nm 6.182    3.008

*Specs*
CPU: 2x Intel(R) Xeon(R) CPU X5675 @ 3.07GHz (24 Threads)
GPU: Matrox Graphics, Inc. MGA G200eW WPCM450 (rev 0a) NVIDIA Corporation
GF106GL [Quadro 2000] (rev a1) -- (
http://www.nvidia.com/object/product-quadro-2000-us.html)
Cuda compilation tools, release 4.2.9, V0.2.1221
(cudatoolkit_4.2.9_linux_64, gpucomputingsdk_4.2.9)
OpenMM 4.1.1-Linux64 (Is this a problem? Page says: the current Gromacs-GPU
release is compatible with OpenMM version 2.0)
Gromacs 4.5.5 (Compiled from source following directions at the above page)
Ubuntu 12.04

I do have to use the option: mdrun_gpu -device "OpenMM: force-device=yes" in
order for Gromacs to run.

If I'm doing things right, then I guess the bad performance is due to my
"incompatible" card. My question is if this will change/improve with
Gromacs 4.6?

Thanks!

--
Mauricio Carrillo Tripp

Biofísica Computacional - Bioinformática

    LAboratorio Nacional de GEnómica para la BIOdiversidad

    Centro de INVestigación y ESTudios AVanzados
    Km 9.6 Libramiento Norte Carretera a León
    36821 Irapuato, Gto., México
    Tel. oficina: (+52) 462 166-3019