[Trilinos-Users] Out-of-core matrices on CUDA

Max Grossman jmg3 at rice.edu
Tue Nov 17 23:10:36 EST 2015


Hi all,

I have one quick clarification about Trilinos on CUDA. It seems that running Tpetra SpMV on CUDA does not at the moment natively support matrices larger than device memory. To gain this capability, I will simply need to tile my CSR matrix myself. Correct? I observe correct behavior up until my matrices start being larger than my GPU memory, then illegal memory access exceptions.

Thanks,

Max


More information about the Trilinos-Users mailing list