Hi all,
At Wageningen UR we have 6 XeonPhis to support our Bionano Irys (optical
mapping). Last couple of days I've been experimenting with the system to
see if I can run my pyOpenCL application on these cards.
I installed mpss 3.3 and opencl 14.2 (newer versions apparently have no
support for the XeonPhi), numpy, scipy and of course pyopencl (python
2.7). It runs on Centos 7, most recent kernel (48 cores, 256GB).
I followed the instructions for mpss and opencl, which worked out
nicely. My application now runs on both the host CPU and the accelerators.
Together with the TU Delft we developed a coarse grained version (thread
& memory wise) for a CPU and a fine grained version for a GPU. The CPU
version runs fine on the host CPU, but crashes on the xeonphi. The
GPU-version runs smoothly on the xeonphi. Not sure yet why the CPU
version crashes, but I'll look in to that.
So I'm very pleased with the results, as I'm now able to use many more
cores for my research. Although in theory the CPU version should run
faster than the GPU version, for now I'm sticking to this set-up. In the
coming weeks I'll benchmark the xeonphis based on my research data and
compare them to Intel CPUs and nvidia GTX/tesla cards.
Cheers,
Sven
_______________________________________________
PyOpenCL mailing list
PyOpenCL@tiker.net
http://lists.tiker.net/listinfo/pyopencl
