Comments on: Nvidia Puts Tesla K20 GPU Coprocessor Through its Paces http://insidehpc.com/2012/09/19/nvidia-puts-tesla-k20-gpu-coprocessor-through-its-paces/ HPC News Without the Noise for Supercomputing Professionals | insideHPC Sun, 19 May 2013 10:46:54 +0000 hourly 1 http://wordpress.org/?v=3.1.1 By: George Millington http://insidehpc.com/2012/09/19/nvidia-puts-tesla-k20-gpu-coprocessor-through-its-paces/#comment-378880 George Millington Wed, 19 Sep 2012 22:53:13 +0000 http://insidehpc.com/?p=31669#comment-378880 Hi Brian, thank you for the question. The performance using all 16 cores of each CPU without a GPU is comparable to using 1 core with 1 GPU per node, so the performance gain in this example is entirely attributable to the Hyper-Q technology. Hope this helps! George NVIDIA Hi Brian, thank you for the question.

The performance using all 16 cores of each CPU without a GPU is comparable to using 1 core with 1 GPU per node, so the performance gain in this example is entirely attributable to the Hyper-Q technology.

Hope this helps!

George
NVIDIA

]]> By: Brian http://insidehpc.com/2012/09/19/nvidia-puts-tesla-k20-gpu-coprocessor-through-its-paces/#comment-378872 Brian Wed, 19 Sep 2012 13:26:13 +0000 http://insidehpc.com/?p=31669#comment-378872 Does the speedup on CP2K perhaps have something to do with the fact that they are utilizing all 16 cores per node, rather than just 1? If CP2K is doing work on both the CPU and GPU, then these results are quite misleading. Another way of looking at this: they used 16 times the CPU resources to get a 2.5X speedup. Does the speedup on CP2K perhaps have something to do with the fact that they are utilizing all 16 cores per node, rather than just 1? If CP2K is doing work on both the CPU and GPU, then these results are quite misleading. Another way of looking at this: they used 16 times the CPU resources to get a 2.5X speedup.

]]>