Is there any advantage of having 2-Way SLI from CUDA standpoint? Is it better to get, say one 256-core card vs two 128-core cards and set them up with SLI?
That's rather old thread, but the main point is still valid: CUDA practically has nothing to do with SLI. On the other side, depending on what you are intending to do, there were some improvements with recent toolkit releases in handling multiple GPUs - for example, CUDA 4.0 introduced so-called Uniform Virtual Addressing, that makes it possible for one GPU to access memory attached to other GPU, without going through main memory for this purpose, which is sort of a step in the direction of NUMA-like architecture.
SLI adds more horse power with additional GPUs but the app needs to be written to utilize them. The current most powerful single-core GPU from NVIDIA has 512 internals cores and it is better(*) than multiple of it on different cards.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.