Hi reading support of C++ AMP on nsight 3.0: http://blogs.msdn.com/b/nativeconcurrency/archive/2013/02/06/remote-gpu-debugging-on-nvidia-hardware.aspx
I have to say it’s very Interesting!
some questions about limitations this post shows:
*it’s says in blog
“For debugging purposes, because of current limitations in the driver, the remote machine must be configured with only one GPU card enabled.”
seems system can’t have more than one GPU means of Nvidia driver or WDDM driver model 1.2 i.e. is going to be fixed by new Nvidia driver or needs need Microsoft OS with revised WDDM to fix this?
*Also seems Nsight 3.0 have some sort of emulation of GPU preemption so I can debug GLSL shaders and DX ones with only one GPU and without remote debugging… what’s holding that for AMP debugging is in fact Microsoft GPU debugging APIs for interop with IHVs or Nvidia limitation now?
Concluding I would want to say for future would be very interesting for C++ AMP debugging at least if even without preemption emulation at least is possible to use only one (local) system altough requiring 2 GPUs, the debugging one as headless display for not hanging display while debugging and the other for display… that’s also have been possible in Nsight for debugging CUDA code for more than a year already I think…