I would like to ask how I could calculate the effective global memory bandwidth for GPUs with compute capacity lower than 1.2.
For example, I have a GTX275 that has 1.3 compute capacity, so the CUDA profiler 2.3 gives the read and write global memory bandwidth in the summary table. However, when I use a 8600GT that has a 1.1 compute capacity, the bandwidth data are not available. Is there any way that I could use the coalesced and uncoalesced counters to manually calculate the read and write bandwidth?
Any suggestions are most welcome.