From: Lucas (lucasbleicher_at_gmail.com)
Date: Tue Aug 13 2013 - 13:06:06 CDT
I've seen this error message in some old posts in the list archives, but
the solutions do not seem to work for me. When trying to run NAMD on a
NVIDIA Quadro 600 in a single machine (running Red Hat Enterprise Linux
6.4), with the command "./namd2 +p2 +idlepoll production.conf >
production.log" I get the following error:
------------- Processor 6 Exiting: Called CmiAbort ------------
Reason: FATAL ERROR: CUDA error on Pe 1 (server device 0): All CUDA devices
are in prohibited mode, of compute capability 1.0, or otherwise unusable.
Segmentation fault (core dumped)
Running "nvidia-smi -q" returns that Compute Mode is Default, and the
model has compute capability 2.1, so I really don't know what may be wrong.
I tried running some of those sample programs at
/usr/local/cuda-5.5/samples/ and they seem to be working fine. "matrixMul"
for example returns:
[Matrix Multiply Using CUDA] - Starting...
GPU Device 0: "Quadro 600" with compute capability 2.1
Computing result using CUDA Kernel...
Performance= 26.76 GFlop/s, Time= 4.899 msec, Size= 131072000 Ops,
WorkgroupSize= 1024 threads/block
Checking computed result for correctness: Result = PASS
Note: For peak performance, please refer to the matrixMulCUBLAS example.
So it seems the system recognizes the card and can run programs on it, but
something is preventing NAMD to do it. What may be wrong?
This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:23:34 CST