Date: Wed, 11 Apr 2012 06:46:04 +0200 From: Lukas Odzioba <lukas.odzioba@...il.com> To: john-dev@...ts.openwall.com Subject: Cuda for nV OpenCL for AMD? Now we are in the situation that Johns OpenCL build "supports" nvidia and amd cards. This leads to problem that opencl kernels are much slower than cuda code for nvidia's cards. To get rid of it we would have to write separate OpenCL kernels for AMD, nVidia and possibly CPU in the future. Can't we just make all cuda formats available for amd devices, and same thing for nvidias? Then we will have to create/manage 2 versions of each format rather than 3. Of course some could say "hey what with users with nv and amd in one box?" 1) They can run two separate instances. 2) Maybe it wont be difficult to create MPI version that interconnects multiple builds? 3) Is this situation common? I think that we should not care about other CUDA capable devices other than GPU's unless they won't be on market (If they ever will be) Do you see any other pros and cons with this "divorce"? Lukas
Powered by blists - more mailing lists
Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.