0
After looking up several benchmark tests that measure the impact different PCI-Express setups have on graphics card performance, I was pretty surprised. It seems almost across the board, halving the lanes from 16 to 8 as well as testing between different generations has an insignificant impact. I am trying to understand why.
I understand that a large bottleneck for running computations on a GPU is copying memory to and from the GPU's VRAM. How then is it possible that doubling the number of lanes does barely a thing? My guesses are card drivers are locked to using up to 8 lanes; My assumption that copy time is significant when compared to the actual computation is incorrect; and/or there are a lot more constraints I don't know about (which is obviously true).
A quick google search will get you to several seemingly credible benchmarks.
Here are the top 2 from my search:
https://www.pugetsystems.com/labs/articles/Impact-of-PCI-E-Speed-on-Gaming-Performance-518
http://www.gamersnexus.net/guides/2488-pci-e-3-x8-vs-x16-performance-impact-on-gpus
I've seen a lot of posts and articles clearly indicating the lack of a difference between setups, but none have explained why at an acceptable level.
These benchmarks all measured gaming and rendering, not computation. With gaming, the textures and such are already on the GPU and the results aren't needed off the GPU. – David Schwartz – 2016-08-24T10:37:15.790
@David I think that answers it then. Thanks for this clarification – Dan – 2016-08-25T00:28:30.247