I was comparing the list of OpenCL devs in Appendix A of the OpenCL Programmers Guide with a page showing [post=“0”]Cuda-Enabled GPU’s[/post].
I did not see any Cuda products, except ION, that were in Cuda, but not OpenCL. The difference in formats makes this difficult. What is the status of the 8100 mGPU for OpenCL? The appendix shows it, but no other data on the line.
It is not important that every Cuda device support OpenCL, but in order allow for the user of a OpenCL program to be different from the programmer (commercial or internal appl), how it fails is important for support purposes. Is it just never found, set to un-available, or blow up when doing something like compiling a kernel? Does anyone have any info in this area?
All GPUs that support CUDA should support OpenCL. ION should work.
OpenCL has lots of functionality for querying available devices.
On systems that don’t have a OpenCL-compatible driver installed, the application will probably fail loading the OpenCL DLL (you can dynamically load the DLL to detect this case).
Thanks,
This was prompted because a different vendor’s beta has an issue where some of their devices get found by clGetDeviceIDs, & clGetDeviceInfo states they are available, but they are not OpenCL capable. This means these devices can get passed any device checking code, only to fail later for unclear reasons. It could make support a lot messier than it needs to be. They acknowledge the issue.
I was just trying to do some checking, using documents, to see if it was possible that NVidia could also have this issue. Finding a device that can run Cuda but not OpenCL was my test. This exercise is complete, given that there are none.
Yes, I am dynamically loading the DLL to detect if OpenCL is present.