Compared to my last post, this should be an easy one. I'm looking at converting that CUDA app to OpenCL for ATI, but looking at the atiopencl example app in the BOINC source code, I'm confused. The standard way to handle multiple GPUs in BOINC with CUDA is to pass in a --device parameter and select the given device number. I don't see any such parameter in atiopencl.cpp.

So how is the case of multiple GPUs supposed to be handled? Should I create one thread per compatible GPU? Should I just ignore the problem? Or did I miss something?

Thanks!

Found it! Apparently the --device parameter is common to all GPU apps.

Perhaps I should file a bug report for the example atiopencl app.

Or email the person writing the OpenCL documentation. See your PM. :)

---