That was kind of my point... He was talking using chiplets to increase the CU count more than what we see now. Which means memory contention. It would be odd to think they would do all that effort with interconnects to put 2x 20CU chiplets together to make a 5700xt. They would put 2x 40 or 4 x20, or w/e, which means it would be fighting for memory the entire time. I think AMD has a pretty good idea on chiplet pros and cons and if they thought it would help them they'd be jumping on it. Also, even with a fast interconnect, they still have latency issues (refer to example of 3100 vs 3300x at same speeds). It isn't an automatic win by just splitting it up into smaller parts. Someone still needs to do the interconnecting of said parts. So if you have great yields at 7nm, and move to chiplet design with slightly better yields (given the same GPU design specs) you may win out, you may break even or you may end up costing even more logistically, but in any case you end up with a slightly slower product. If you go to a huge die and can get a real benefit to yields, then it could be a win if you up your memory bandwidth to match the increase in cores.or.xome up with some large high speed cache near the GPU that helps mask the latency issues (which I know they are researching this now).Nope. Traditional multi GPU even when implemented properly is fundamentally flawed due to the alternate frame rendering process. It causes input lag and all sorts of problems.
Even if you get 100% scaling (which I have never seen) it is still pretty bad.
The way 3DFX originally did it by alternating scan lines was a much better approach, but they never solved the scaling issue. Later development of Split Frame Rendering got closer to solving this issue but because people didn't demand it, and AFR was easier to program that's the crap we got.
With a chiplet design, as long as you could get good fast interconnects you wouldn't have any more contention for the memory than you would by adding more cores to a very large single GPU, because it would be a very large single GPU, just split across multiple dies.
We wouldn't have to worry about multi-GPU at all in game, because it would be one GPU.