AW: vmd-l: NAMD 2.11 Cuda does not work with Quadro FX 3800

From: Greipel.Joachim_at_mh-hannover.de
Date: Fri Mar 11 2016 - 01:47:27 CST

Updating the driver is not an option: the Nvidia driver is 340.96. This is
the newest driver that supports the FX3800. Drivers of the 352.x series do
not support the cards.

On the system with the two Tesla K20 Cards, which is working, the driver
that is installed is 340.93. So the older driver with newer cards work.

 

Joachim Greipel

 

 

Von: Norman Geist [mailto:norman.geist_at_uni-greifswald.de]
Gesendet: Donnerstag, 10. März 2016 23:44
An: Greipel, Joachim Dr.
Cc: namd-l_at_ks.uiuc.edu
Betreff: AW: vmd-l: NAMD 2.11 Cuda does not work with Quadro FX 3800

 

I would suggest updating the Nvidia driver as 1st thing to try.

 

Norman Geist

 

Von: owner-vmd-l_at_ks.uiuc.edu [mailto:owner-vmd-l_at_ks.uiuc.edu] Im Auftrag von
Greipel.Joachim_at_mh-hannover.de
Gesendet: Donnerstag, 10. März 2016 08:57
An: vmd-l_at_ks.uiuc.edu
Betreff: vmd-l: NAMD 2.11 Cuda does not work with Quadro FX 3800

 

Hi,

 

yesterday I updated NAMD from 2.9 to 2.11. Unfortunately I got the following
Error when trying to run the CUDA enabled Binary on a System with two Quadro
FX3800 graphics cards:

 

FATAL ERROR: CUDA error binding lj table to texture on Pe 2 (ceo-01 device
0): invalid texture reference

[2] Stack Traceback:

  [2:0] CmiAbort+0x41 [0xefde44]

  [2:1] _Z8NAMD_diePKc+0x93 [0x64a6c3]

  [2:2] _Z13cuda_errcheckPKc+0xec [0x826b6c]

  [2:3] _Z22build_cuda_force_tablev+0x446 [0x8176e6]

  [2:4] _Z15_processHandlerPvP11CkCoreState+0x727 [0xd95e07]

  [2:5] CsdScheduler+0x5ff [0xf0722f]

  [2:6] [0xefdb90]

  [2:7] [0xefb293]

  [2:8] +0x7dc5 [0x7fa183f10dc5]

  [2:9] clone+0x6d [0x7fa18301228d]

[2] Stack Traceback:

  [2:0] LrtsAbort+0x70 [0xefe524]

  [2:1] CmiAbortHelper+0x102 [0xefddfd]

  [2:2] CmiAbort+0x41 [0xefde44]

  [2:3] _Z8NAMD_diePKc+0x93 [0x64a6c3]

  [2:4] _Z13cuda_errcheckPKc+0xec [0x826b6c]

  [2:5] _Z22build_cuda_force_tablev+0x446 [0x8176e6]

  [2:6] _Z15_processHandlerPvP11CkCoreState+0x727 [0xd95e07]

  [2:7] CsdScheduler+0x5ff [0xf0722f]

  [2:8] [0xefdb90]

  [2:9] [0xefb293]

  [2:10] +0x7dc5 [0x7fa183f10dc5]

  [2:11] clone+0x6d [0x7fa18301228d]

 

Until NAMD 2.9 everything used to work fine.

Testing of NAMD 2.11 with CUDA on a system with two Tesla K20 Cards also
works.

 

System: Linux CentOS 7.2, Nvidia Driver 340.96,

Commandline for Testing (Lysozym-Tutorial): namd2 +idlepoll +p4
ubq_ws_eq.conf > ubq_ws_eq.p4.gpu.log

Graphics Cards: Two Nvidia Quadro FX3800

 

Something must have changed from 2.9 to 2.11 in CUDA Support? Someone any
hints?

 

Regards,

Joachim

 

-- 
Dr. rer. nat. Joachim Greipel 
 
Med. Hochschule Hannover 
Biophysikalische Chemie OE 4350 
Carl-Neuberg-Str. 1 
30625 Hannover 
Germany 
Fon: +49-511-532-3718 
Fax: +49-511-532-16-3718
E-Mail: greipel.joachim_at_mh-hannover.de
 

This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:21:51 CST