HPCG for Nvidia Tesla platform

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello, I was wondering if there is a Nvidia library optimized High-Performance Conjugate-Gradient (HPCG) implementation available for downloading. Target is a Haswell cluster of K80 GPUs.

Thanks!
Michael

2

The binary is available at http://www.hpcg-benchmark.org/software/index.html

3

Thanks for the pointer drM … Does it adapt its run-time behavior based on the total number of available GPUs per node?

Best regards
Michael

4

Yes, you schedule one MPI rank per GPU and the code automatically assigns one.

For example, to run on two nodes with 4 GPUs each, a similar command should be specified:
mpirun -np 8 ./xhpcg_cuda6.5_openmpi

5

Does there exist a version for the tegra k1 board, too?
Unfortunately, the binary x86 version is not compatible with this arm board :(

6

There is no public binary for ARM.

7

Thank you for your answer.

If its not public available, is there a way to get it non-public for internal use only?