Is the CUDA Occupancy Calculator provided by NVIDIA out of date?
The CUDA Wikipedia page lists the "Maximum number of threads per block" as 1024 for compute capability 2.x.
Is开发者_如何学Go this information incorrect as I have read elsewhere that 512 is always the max number of threads per block? Also the CUDA occupancy calculator won't allow more than 512 for the "Threads Per Block" setting.
Edit: The Occupancy Calculator I was looking at was out of date. The newest version can be found on this page.
There is a newer version available. Go to the CUDA 4.0 download page, and select "Documentation". Scroll down to find the CUDA Occupancy Calculator v. 2.4 (yours is 2.1, right?)
http://developer.download.nvidia.com/compute/DevZone/docs/html/C/tools/CUDA_Occupancy_Calculator.xls
Here is a direct link to save you the effort.
精彩评论