4 ms·
> Word counting is primarily IO-bound, and it is much too expensive to ferry the file contents all the way to the GPU over the (relatively) slow PCI Express bus
by e12e 7y ago
> Word counting is primarily IO-bound, and it is much too expensive to ferry the file contents all the way to the GPU over the (relatively) slow PCI Express bus just to do a relatively meagre amount of computation.
After seeing that it's possible to play crysis using software rendering on an AMD Rome cpu with 128 hw threads [1] - might this lead to some vindication for AMD sticking with opencl (assuming exposing such a cpu via opencl) - or is it just simpler to ignore that (in general and for futark) and just use regular threads for parallelizing aacross many cpu cores?
[1]
https://news.ycombinator.com/item?id=21339652 https://news.ycombinator.com/item?id=21339652
- Mathnerd314 7y agoCUDA is proprietary to NVidia, and is pretty much the standard for GPU computing. AMD's been chipping away with OpenCL, Vulkan/GLSL, https://github.com/RadeonOpenCompute/hcc/wiki https://github.com/RadeonOpenCompute/hcc/wiki, etc. but not much luck so far. I wouldn't say AMD's been "sticking with" OpenCL, if anything it seems like they will deprecate it in a few years, as the plan is to fold OpenCL into Vulkan. I guess it is possible to use OpenCL on the CPU as well, but it seems to be intended mostly for testing purposes. The Crysis software renderer uses threads: https://github.com/google/swiftshader/blob/master/src/Common/Thread.hpp https://github.com/google/swiftshader/blob/master/src/Common...
- rrss 7y agoLast I checked, AMD implemented the cuda apis as "hip."