Are they? Can openCL programs be debugged on the CPU and then executed on the CPU or GPU? Is there better support? Anything negative? Let me know, thanks.
The OS X OpenCL implementation for NVIDIA GPUs has significant performance overheads in 10.6.2, compared with the same hardware used on Linux or Windows.
I would say it’s too immature for serious development. We are currently pushing CUDA on OS X (which performs well and equally with other platforms) and OpenCL on Linux and Windows. Since there is a CPU implementation from AMD for Linux and Windows, and the GPU NVIDIA driver supports printf-style debugging, both of those platforms are better options at this time.
printf-style debugging on Windows?
Can you tell me how you managed this to work?
I using GPU Computing SDK 3.0 Beta as well as the corresponding drivers and it doesn’t work. Neither does it work with ATI Stream 2.0.1 on CPU.
Tried both on Windows 7 64-Bit (Nvidia 8500 GT).
I also tried printf in kernel on my MacBook and there it works without problems.
What I can say is Mac is the most stable right now. If I have something weird go wrong, my first reaction of trying it on OSX has paid off more than once.