Posts Tagged ‘ gpu ’

Josh Simons

SC11: Virtualized GPU Compute

November 21, 2011
By
SC11: Virtualized GPU Compute

  My friend Cydney brought me to the AMD booth and introduced me to Scot Schultz, pictured here. He was showing a nice demo of an N-body simulation written in OpenCL running in a Linux VM on vSphere 5.   The GPU was made visible to the VM using VMware Direct Path I/O in the same manner we used internally for running CUDA tests on both AMD and nVidia GPUs. As reported here , we achieved close to 98% of native performance for most tests we tried.   It was excellent to see AMD showing off the ability to achieve high compute performance with GPUs in a virtual environment!

Read more

Josh Simons

GPU Computing (GPGPU) in a Virtual Environment

October 5, 2011
By

For you GPGPU fans out there, Lawrence Spracklen from our Performance Engineering team has been running some CUDA benchmarks on vSphere and has written up his results on the VROOM! performance blog. Short summary: We generally see about 98% of native performance on the tests done thusfar. Check out the details here .

Read more

Josh Simons

New Chinese Supercomputer Poised to Take #1 Slot on TOP500

October 27, 2010
By

The New York Times reports that the Chinese are upping the ante significantly with a new supercomputer , which is predicted to take the #1 slot on the HPC TOP500 list when the latest version is announced at SC ’10 in New Orleans in a few weeks. Oak Ridge National Laboratory’s Jaguar   system is ranked #1 on the current list, which was released in June. The Chinese system, the Tianhe-1A, is rumored to thump Jaguar’s peak performance by an impressive 1.4X. The system apparently makes heavy use of nVidia GPUs, as described in this nVidia press release .

Read more