[ofa-general] strange problem with infiniband and GPUs

[Brian Budge]

Hi all -

I'm new to the list, and I hope this is the correct place to post this.  I
am running an MPI application which uses CUDA and NVIDIA GPUs to accelerate
computation.  I am using mvapich2 to get multi-thread-safe MPI with
infiniband.

If I run mvapich2 configured for tcp, my application runs fine (or if I run
it in single node mode without MPI), but if I run it configured for
infiniband, my application fails on GPU initialization about 80% of the time
(the other 20% of the time, my application runs fine to completion).  I'm
not sure what could be happening.

I'm not sure if somehow one of the infiniband drivers could be interacting
with the nvidia driver?

Thanks for any help,
  Brian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20080109/86276746/attachment.html>