Originally Posted by Coydog
Ok, thanks for clearing this up for me. I don't deal with SLI/CF at all so never really learned about it. I think I get it now. Basically in a SLI/CF setup, 100% =/= the total load each GPU can assume, basically in a SLI/CF setup only a certain percent of normal GPU load is utilized between the 2 cards for use in rendering the game onto the display? Again, want to make sure I get it right.
Let's take some example. Let's say you run a game with 1 GPU and you get 30 FPS. You run the same game with 2 GPUs and you get 50 FPS. This means that adding a second card gave you 20 fps (whoch represents 66% of the 30 FPS gained by the first card). In this case, for this game, the scalling is 66%
Now both cards are running at 100% GPU clock. So why don't you get double performance then? Does this mean the second card runs at 66% ? Well no ....for 2 reasons:
.1. The GPUs have to communicate with each other. Handling data that comes from the other GPU takes time ; and that is the time you won't be doing what you "should do" - rendering
.2. The tasks are somethimes difficult to split equally. Imagine you have one 2l bottle of water and 1l bottle of water. In total you have 3l. So if you want split the water between 2 guys, you would need 1.5l each. But since you can give ony one bottle to each, you do not divide the water equaly.
Point 1 is taken care by the drivers. Point 2 is taken care by the game developers. That's why you will have games with different GPU scalling - as it devepends on how good ther drivers / game developers are. If your Crossfire / SLI doesn't work, it means you don't have a driver or the developer(s) wrote the source code for a single GPU -e.g. most console ports
Now, if you want technical data about the GPU intercomunication (e.g. scan interleave),protocol used, workload split , memory bus handling, ...etc you probably have to do your own research , as I have no idea how that stuff works Edited by WanWhiteWolf - 6/1/16 at 5:29pm