By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Pemalite said:

Correct.

Simple reason for this, PCI-E and the SLI/Crossfire bridge doesn't have the bandwidth or latency required in order to share the memory contents on each card for one big fat giant memory pool.
However, if you go farther back in the cache hierachy, aka System Ram, the GPU's can all share the same memory before it gets sent off to each GPU, which is what is important for compute tasks.

However, with that said despite the memory pools having duplicated information, SLI/Crossfire still manages almost 100% scaling in dual GPU gaming scenarios, compute is a whole different ball game however.

So in the end, even though I have in total 9Gb of GDDR5 memory, only 3Gb of that is useable by games, the whole 9Gb however is available for compute. :)

Are you saying if I had two Titan with 6GB GDDR5 each they will use only 3GB for games? I thought it was 6GB (from the 12GB combined).

Thanks BTW.