I was going through the specification document for GM 240. For some reason I m not able to find the bandwidth specification (gb/s).
I do see memory bandwidth but thats for the memory out side of chip.
Any document available on this?
I was going through the specification document for GM 240. For some reason I m not able to find the bandwidth specification (gb/s).
I do see memory bandwidth but thats for the memory out side of chip.
Any document available on this?
I’m not aware of a maxwell GM 240. There is a maxwell GM 204 chip. Also not sure what you mean by “the specification document”. I suppose that might refer to several things.
In any event, NVIDIA does not publish (AFAIK) bandwidth specs for any of the caches: L1/Tex, L2, (for any GPU products, to my knowledge).
These numbers generally have to be arrived at via some form of reverse engineering, micro-benchmarking, or the like, or else arrived at anecdotally by culling presentations and other related collateral for side-references.
Shared memory (which is not a cache) bandwidth can be inferred from available documentation/specifications, e.g. for Fermi and Kepler:
[url]cuda - Shared memory bandwidth Fermi vs Kepler GPU - Stack Overflow
Similar analysis should be usable on Maxwell. (Note that Maxwell dropped the 64-bit-bank mode that Kepler had).