Re: Accelerated MD

From: Jeff Wereszczynski (jmweresz_at_mccammon.ucsd.edu)
Date: Wed Aug 29 2012 - 23:19:39 CDT

Ah, you are using CUDA! Yes, that doesn't surprise me that the two are
incompatible. Perhaps the developers could add a warning so that others
don't do the same thing?

On Wednesday, August 29, 2012, Thomas Albers wrote:

> Dear Jeff,
>
> > Yes, something looks weird there.
>
> It seems there is something very wrong when CUDA acceleration is
> combined with a boost to the total potential:
>
> Here's the relevant output for aMDd mode with graphics card acceleration:
>
> Info: NAMD 2.9 for Linux-x86_64-ibverbs-CUDA
> ...
> Info: BOOSTING DIHEDRAL POTENTIAL
> Info: accelMDE: 6900 KCAL/MOL, accelMDalpha: 570 KCAL/MOL
> Info: accelMD WILL BE DONE FROM STEP 0 TO THE END OF THE SIMULATION
> Info: accelMD OUTPUT FREQUENCY 100
> ...
> ACCELERATED MD: STEP 20000 dV 253.844 dVAVG 275.381 BOND 1338.24 ANGLE
> 6925.25 D
> IHED 6372.08 IMPRP 69.7927 ELECT -7868.19 VDW -3003.61 POTENTIAL 3833.56
> ...
> ETITLE: TS BOND ANGLE DIHED IMPRP
> ELECT VDW BOUNDARY MISC KINETIC
> TOTAL TEMP POTENTIAL TOTAL3 TEMPAVG
> PRESSURE GPRESSURE VOLUME PRESSAVG GPRESSAVG
>
> ENERGY: 20000 1338.2441 6925.2481 6372.0787 69.7927
> -75197.2498 1439.0778 0.0000 0.0000
> 18897.7076 -40155.1008 300.7371 -59052.8084
> -40031.7866 300.6322 -612.8711 -603.8231
> 281748.7794 -7.7389 -7.5624
>
> and here the equivalent output with plain NAMD:
>
> Info: NAMD 2.9 for Linux-x86_64-ibverbs
> ...
> ACCELERATED MD: STEP 6000 dV 333.131 dVAVG 323.749 BOND 1253.58 ANGLE
> 7059.32 DI
> HED 6266.93 IMPRP 70.9914 ELECT -75591.7 VDW 1599.76 POTENTIAL -59341.2
> ...
> ENERGY: 6000 1253.5799 7059.3184 6266.9277
> 70.9914 -75591.7370 1599.7616 0.0000
> 0.0000 18788.4535 -40552.7046 298.9985
> -59341.1581 -40425.8879 299.9954 241.0034
> 256.3487 278915.3123 -8.1178 -8.3545
>
> The numbers are comparable.
>
> Here's output for aMDT mode with GPU acceleration:
> Info: NAMD 2.9 for Linux-x86_64-ibverbs-CUDA
> ...
> Info: ACCELERATED MD ACTIVE
> Info: BOOSTING TOTAL POTENTIAL
> Info: accelMDE: -56550 KCAL/MOL, accelMDalpha: 4690 KCAL/MOL
> ...
> ACCELERATED MD: STEP 500 dV 0 dVAVG 0 BOND 1325.53 ANGLE 6894.1 DIHED
> 4092.79 IMPRP 61.4494 ELECT -7851.73 VDW -2788.71 POTENTIAL 1733.43
> ...
> ENERGY: 500 1325.5256 6894.0969 4092.7913
> 61.4494 -75364.0230 1507.6438 0.0000
> 0.0000 19104.2109 -42378.3050 304.0234
> -61482.5159 -42247.3078 300.6413 -361.8518
> -328.5936 280609.2766 20.7882 21.6944
>
> And here is output for the same with vanilla NAMD:
> Info: NAMD 2.9 for Linux-x86_64-ibverbs
> ...
> Info: ACCELERATED MD ACTIVE
> Info: BOOSTING TOTAL POTENTIAL
> Info: accelMDE: -56550 KCAL/MOL, accelMDalpha: 4690 KCAL/MOL
> ...
> ACCELERATED MD: STEP 3000 dV 21.1076 dVAVG 17.9464 BOND 1476.16 ANGLE
> 7939.6 DIHED 4252.22 IMPRP 69.9781 ELECT -71794.8 VDW 1181.47
> POTENTIAL -56875.4
> ...
> ENERGY: 3000 1476.1563 7939.5996 4252.2218
> 69.9781 -71794.7873 1181.4665 0.0000
> 0.0000 19102.0412 -37773.3239 303.9889
> -56875.3651 -37660.1795 301.9512 -103.8260
> -80.9638 288694.1023 -11.3759 -11.3094
>
> It would seem that with the CUDA version, no boost is applied to the
> total energy.
>
> Thomas
>

-- 
Jeff Wereszczynski
Postdoctoral Scholar
University of California, San Diego
http://mccammon.ucsd.edu/~jwereszc

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:59 CST