Open himcraft opened 3 months ago
Hi, I am also facing a similar issue. I tried to run the EB-C14 compression ramp case in a dedicated GPU cluster by setting CUDA and MPI flags to TRUE. I kept nprocs=16, ngpu=1 thereby np=16 The case fails to start, reporting an out of memory error by AMReX_Arena.cpp; and an error generated by the same line 749 in AMReX_GpuLaunchFunctsG.H file
The first issue, i.e. out of memory error caused by number of MPI processes, goes off once I adjust the np count to be same as ngpus. However, the second error, reported by the AMReX_GpuLaunchFunctsG.H file, is still there. Requesting for help!
It is probably due to cuda version IMO.
When I switched to 12.6 instead of 12.2, the error disappeared. The same error occurred when I recently tested on another HPC system using cuda 12.2.
Hello. Recently I wanted to run the PMF case with my laptop GPU, I changed
USE_CUDA
toTRUE
inGNUmakefile
and recompiled following the instruction on the document. While when I run./PeleC3d.gnu.CUDA.ex pmf-lidryer-cvode.inp
, it promptedThe contents in
Backtrace.0
areThe CUDA version is 12.1. Could it be my CUDA driver is installed wrongly? While I can run my own
.cu
code without error.Thanks in advance.