Posted November 21, 201212 yr Got my head around some C++ AMP at the weekend and wrote a benchmark for DX11 GPUs that probes some of the computational aspects in grid solvers. Each stage lasts for 30 seconds - CPU score is designed to be less than 5% of the overall grade, but as always having a fast CPU helps keep the GPU fed. AMD doesn't fair too well, but that's more a weakness of the DX11 API with C++ AMP for this sort of thing. Download below and feel free to test! // System Compute // (c) 2012 Ian Cutress // // System Compute is a benchmark designed to stress the CPU and AMP device of the system // // Changelog // // v0.4 : Added CPU detection // v0.3 : Added AMP MatMul // v0.2 : Fixed Score calculation // : Added 2D IMP CPU // : Made it look nicer // v0.1 : Initial Build // : 2D-CPU // : 3D-CPU // : 2D-AMP // : 3D-AMP Still need to add result verification tools. SystemComputev0.4.zip Edited November 21, 201212 yr by borandi
November 21, 201212 yr anyway to get the source code? i have a buddy who is interested in coding part of benchmarking..so was thinking if anyway i can get him to help out.
November 21, 201212 yr Author Code is not public, same reason FM doesn't make its code public I suggest getting a book or two ofc. I get a portion of my computational code from back when I did my PhD, as such it's based on real life problems. As long as your buddy is proficient in C++, have a look at n-body simulations. Those are always interesting and taxing, or black-scholes for finance. It's a case of picking up an algorithm and having a go. A strong background in mathematics always helps, especially when writing optimised code :thumb:
November 21, 201212 yr actually i was more interested on why AMD card is not fairing well...wanted to look into that part
November 23, 201212 yr Author actually i was more interested on why AMD card is not fairing well...wanted to look into that part AMD GPUs fair OK, but a big driver of single GPU perf is single core speed as everyone knows The Trinity A10-5800K CPUs are about 35% off an Ivy i3-3225 at stock, hence the big decrease. It just so happens that PCIe speed is all critical as well. PCIe 3 > PCIe 2, and as much BCLK as you can fire at it also seems to work. http://www.overclock.net/t/1329409/i-made-a-benchmark-systemcompute-v0-4/0_100#post_18651666 is where the main thread is
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.