Originally Posted by rluker5
I've thought for a while that you can't just present a bunch of chiplets to a game and have the drivers just be able to make sense of what to do in any reasonable amount of time. Different games ask different things of gpus and there isn't an easy way to just split the load by some quick pattern over a satisfactory percentage of them.
Kinda. People forget how well the muilti-GPU Voodoos fared after 3DFX died and there wasn't any more driver updates to aid running newer games. Difference between those mGPU cards and the more notable mGPU tech we all think of with modern CFX/SLI is that the Voodoo was designed from the ground up to be multi-chip with the VSA-100 designed to go up to 32 chips, kinda like a chiplet is while modern mGPU setups are kinda just plugging two GPUs into a PC and trying to make things work with software.
Hell, even modern mGPU works decently provided it's getting good driver support. The main reason it's struggled so much and is dying off now isn't because it doesn't work, it's because it costs so much to make it work relative to how many gamers actually used it and how much of a difference it made when a single high-end GPU is already pushing high FPS anyway. Give it the right market and it'd take off again, it'd just have to be a market that gets it into as many hands as possible to justify the dev work needed. (I can see it being a thing in Ryzen laptops with a Navi dGPU; basically iGPU or dGPU depending on load when using the battery, but run mGPU with both while plugged in for maximum performance. I mean, hybrid CFX mostly worked well for the few months it had proper driver support...Just was kinda pointless in a desktop.)