I’m wondering where I can find hardware specification for different versions of card that supports CUDA (say, GTS 8600 or GT200). I’m interested in the number of MPs, warp size, and global synchronization latency. By the way, I assume global synchronization latency is a constant, # of blocks > # MPs?
Except that Nvidia marketing does not care much for consistency between model names and actual products. At least the 8400 GS exists in two versions, the older (g86) having two multiprocessors just like appendix A says, and the newer (g98) having only a single core, which OTOH is quite a bit faster … I believe the latter - just for the sake of confusing any wannabe expert - is also sometimes known as the 9300 GS?