Originally Posted by brlawyer
After reading a number of nMP benchmarks on BareFeats, I still fail to understand why NVIDIA does not have an agreement with Apple to support SLI
You mean why AMD doesn't have an agreement to support CrossFire. The nMP uses AMD GPUs. It would be SLI if they used NVidia GPUs.
It seems like it would be a trivial thing for them to add. It's really just splitting the required frame and rendering each part separately. It doesn't seem like it would be very difficult to implement. They just load up the memory on each GPU with the same data and then tell GPU 1 to render all the even lines into the framebuffer and then GPU2 to render all the odd lines or it can be left/right parts of the frame and then display the framebuffer.
I don't think they really intend the second GPU to be used like that though. It seems as though the second GPU is intended to be used as a co-processor. When you use a single GPU with OpenCL tasks, the user interface lags because it has to try rendering on the GPU too. I think this is what happens with Final Cut Pro X. The timeline is more laggy than it is in other apps.
For apps that can use the two GPUs best together, the developer can take advantage of both.
Sometimes it's not a good idea to use both because it generates more heat. If a game runs at 90FPS maxed out with 2 GPUs and uses say 400W, it's mostly wasted. Running at 45FPS at 200W is better. A single D700 GPU should run every game maxed out above 30FPS, except really poorly optimized games.
Having the option would be nice so people don't have to wait on the app developer to use both.