Anybody here try CUDA programming in OS X?

[struggle4progress]

In particular, is CUDA working in Snow Leopard?

And can CUDA with (say) one GTX280 really speed stuff up much? For example, do the matrix-vector operations in CUDA's version of BLAS really work quickly and well?

[struggle4progress]

1. There's a lot of unhappiness out there. These comments worked for me:

~snip~
* Install the toolkit before the driver. Installing the driver before the toolkit usually works, but occasionally libcuda.dylib will get deleted.
* The compiler fix on Snow Leopard (for now) is along the lines of "disable blocks within CUDA files." This means you can't use Grand Central Dispatch within .cu files.
* 64-bit kernels are unsupported.
~snip~
http://forums.nvidia.com/index.php?showtopic=106394&pid=587641&mode=threaded&start=#entry587641

~snip~
make clobber
make i386=1
~snip~
http://forums.nvidia.com/index.php?showtopic=106394&pid=588131&mode=threaded&start=#entry588131

In particular, I didn't edit the common make file to insert -m32 in the compiler flags. Some of the examples (like nbody) set off persistent system scream-beeps until terminated

[struggle4progress]

2. OK. The scream-beep wasn't from the computer but from the UPS,

which objected to the current load. Apparently using a graphics card to do nbody computations is current intensive.