Compute Capability 1.1 32-bit words??

hi all,

what does mean with “32-bit” words? Is it relevant with OS? For example, i can’t use the atomic funtions with a 1.1 enabled card on a 64-bit os?


It means that all current GPU are 32 bit devices, and the word size is 32 bit. Compute 1.1 devices can be used with 64 bit operating systems, the only different is that pointer sizes change to match the size of host pointers - so the same GPU uses 32 bit pointers when running on a 32bit OS and 64bit pointers when running on a 64bit OS.

got it! thx!

i got the error “identifier “atomicMin” is undefined” currently as i want to use it. I’ve already add

-arch sm_11

in my CUDA Build rule v2.3.0, which I got from here. But I still saw following assertion:

“C:\CUDA\bin64\nvcc.exe” -arch sm_10 -ccbin "C:\Program Files (x86)\Microsoft Visual Studio 9.0\VC\bin …

how can I solve this problem?

Can’t help you with windows, I am sorry. atomicMin should be OK for sm11 on global memory. Shared memory atomics are only supported on sm12 and sm13 hardware.