HPCBIOS_2012-99: GPU Productivity EnvironmentΒΆ

Note

GPU productivity environments have recently become an important part of scientific computing and visualization. The following is an attempt to define which ones are relevant for HPCBIOS compliant sites and set an action plan.

  • BC Policy: HPCBIOS_2012-99
  • Date of Policy: 2012-12-15

The main objective of this policy is to provide the following common GPU productivity environment across all HPCBIOS resources:

Package/Tool Requested version(s) Compliance level Reference
CUDA Toolkit v4.2 or later MUST https://developer.nvidia.com/cuda-downloads
rCUDA v4.0.1 SHOULD http://www.rCUDA.net
PyCUDA 2012.1 SHOULD http://mathema.tician.de/software/pycuda
PyOpenCL 2012.1 SHOULD http://mathema.tician.de/software/pyopencl
MAGMA v1.3 SHOULD http://icl.cs.utk.edu/magma
ViennaCL v1.4.2 SHOULD http://viennacl.sourceforge.net
AMD math library for OpenCL (APPML) v1.10 SHOULD http://developer.amd.com/libraries/appmathlibs/Pages/default.aspx
Rgputools for R language v0.28 SHOULD http://cran.r-project.org/web/packages/gputools
Intel OpenCL SDK v1.5 SHOULD http://software.intel.com/en-us/articles/vcsource-tools-opencl-sdk
CULA & CULASPARSE R16a & S4 SHOULD http://www.culatools.com
VCL v1.19 SHOULD http://www.mosix.org/txt_vcl.html

This GPU productivity environment will be supplemented with other such productivity tools, as they become available on allocated systems.

Potential sources of information for further future upgrades of this list include: