CUDA 2.2 released

[Tim Murray]

and here (http://forums.nvidia.com/index.php?showtopic=96303) is my ridiculously long post full of features, fixes, downloads and the like. Brief summary:

- cuda-gdb for RHEL5 32-bit and 64-bit, but it works on more platforms than that (I've actually used it on Ubuntu 9.04)
- exclusive mode--setting up CUDA/MPI apps on multi-device systems is no longer an exercise in frustration
- zero-copy: read across PCIe within your kernel
- dramatically improved profiler for GT200

Anyway if you're running 2.1 (or looking at 2.1), you should definitely use 2.2 instead.

[Scali]

Yay, just downloaded and installed it.
For a moment I was afraid that G8x support would be dropped in Cuda 2.2 (http://forum.beyond3d.com/showpost.php?p=1285083&postcount=14), but that doesn't seem to be the case. Seems to work just fine on my 8800GTS320 so far.

[DegustatoR]

Didn't want to start a new thread about it:
http://developer.nvidia.com/object/gpucomputing.html
http://developer.download.nvidia.com/compute/cuda/3_0/docs/NVIDIA_FermiCompatibilityGuide.pdf
http://developer.download.nvidia.com/compute/cuda/3_0/docs/NVIDIA_FermiTuningGuide.pdf

[pcchen]

CUDA 3.0 (http://developer.nvidia.com/object/cuda_3_0_downloads.html) is released. Its OpenCL headers seem to be updated to include opencl.h (Khronos now think it's better to have a single header file to include all necessary headers on a system).