RSS
09-13-13, 07:13 PM
We have a veritable smorgasbord of GFN apps available now:CPU:There are three CPU versions of Genefer available for the short GFN tasks. The SSE2 version is the slowest and is sent to 64 bit hosts that don't support SSE3. (Only very early Athlon-64s don't support SSE3). The new SSE3 version of Genefer is about 20% faster and is automatically sent to CPUs that support SSE3 (that's most CPUs). The AVX version of Genefer is about 50% faster than the SSE2 version and is automatically sent to Linux and Mac hosts with CPUs that support AVX, i.e., Sandy Bridge and later CPUs. The server can't detect whether a Windows host supports AVX, so to get the faster GeneferAVX app on Windows you must explicitly tell the server to send it by checking the "Force AVX" box on the PrimeGrid Prefences (http://www.primegrid.com/prefs.php?subset=project) page.The SSE3 discussion is here (http://www.primegrid.com/forum_thread.php?id=5257) and AVX is here (http://www.primegrid.com/forum_thread.php?id=5179).ATI/AMD GPUs:At long last we have an operational OpenCL version of Genefer that run on ATI/AMD GPUs. Double precision floating point hardware is required on the GPU. Discuss OpenCL here (http://www.primegrid.com/forum_thread.php?id=5255), and the discussion about compatible ATI/AMD GPUs is here (http://www.primegrid.com/forum_thread.php?id=5254).Nvidia GPUs:In addition to the CUDA app we've been using, the OpenCL app for ATI/AMD is also available for Nvidia. Use whichever one runs faster on your hardware. In general, it seems as if CUDA is faster on older, pre-Kepler GPUs (GTX 5xx and older), and OpenCL is faster on newer Kepler GPUs (GTX 6xx and newer.)Discuss OpenCL here (http://www.primegrid.com/forum_thread.php?id=5255), and the discussion about compatible Nvidia GPUs is here (http://www.primegrid.com/forum_thread.php?id=3920).
More... (http://www.primegrid.com/forum_thread.php?id=5258)
More... (http://www.primegrid.com/forum_thread.php?id=5258)