Hello,

I would like to report that after recompiling openmpi with cuda support and 
recompiling petsc with the option

>>>> --download-hypre-configure-arguments=--enable-unified-memory

I still got the error

>>>> CUDA ERROR (code = 101, invalid device ordinal) at memory.c:139

when using hypre.

So it is not an issue of mpi cuda support. I ran ex19 without the 
"-use_gpu_aware_mpi 0” option. Maybe it has something to do with the Tesla K40m 
GPUs. However, on their website:

β€œIn CUDA 6, Unified Memory is supported starting with the Kepler GPU 
architecture (Compute Capability 3.0 or higher), on 64-bit Windows 7, 8, and 
Linux operating systems (Kernel 2.6.18+).”

In any case, thanks again for helping out with this.

Juan S.

Reply via email to