AW: GTX-660 Ti benchmark

From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Mon Sep 17 2012 - 01:31:12 CDT

Of course, as already pointed out, and I guess I supposed it some weeks ago already, the pcie bandwidth for the doubled power of the gpus is way too low for namd. But it would be interesting to see with pcie 3.

 

Norman Geist.

 

Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im Auftrag von Wenyu Zhong
Gesendet: Freitag, 14. September 2012 05:02
An: namd-l
Betreff: Re: namd-l: GTX-660 Ti benchmark

 

Hi,

 

I would like to share my results.

The hardware used are I5-3450_at_3.7G, Z75, dual ddr3 1600, and GTX460_at_1.46G or GTX 660Ti_at_0.941G. The apoa1 benchmark is modified to using "rigid all" and "timestep 2fs". Also the downloaded Linux-x86_64-multicore-CUDA builds of NAMD 2.9.

 

Results in ms/step.

 

1*460 2*460 1*660ti

42 32 43

 

The power consumption with i5_at_3.7G+660ti running apoa1 is about 260w.

 

The results are very disappointed.

 

And I have test 660ti with acemd. 12ms/step of apoa1 benchmark is obtained with "timestep 2fs" and "rigid all".

 

Wenyu

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:22:05 CST