I found this on repi's presentation, slide 49:
http://www.slideshare.net/repii/parallel-futures-of-a-game-engine-2478448?from=ss_embed
irrelevant seems to be the keyword here.
Some points you missed:
1. 2015.
2. GPUs of 2010 are in the 2TFLOPs range; do we expect a 25x FLOP increase in less than 5 years?
3. GPU architecture is very far away from running gamecode.
4. CPUs are still lagging back in the 200GFLOP range.
5. Yes, OOOe is irrelevant for significant amounts of FLOPs (especially graphics which when combining programmable FLOPs with non-programmable ones found in TMUs and such is already in the tens of TFLOPs) but ...
6. but ... the game loop, as non-FLOP intensive as it may be can hold you back a lot.
7. And parallelizing it isn't so easy/efficient.
8. Creating a parallel game loop on GPU-like "cores" would be a nightmare. Or even SPEs.
You could increase the list but there is this major factor: Even if DICE, with all their mojo and EA moneyhats wants a system like this, the real question is if it is good for the market.
The market is growing decidedly in the area of mobile and "arcade" like sectors. Likewise a slew of software is done by smaller studios. Not to mention development expectations are increasing while turn around rate has stabilized. There is a
premium on turning games around, on time and on budget, not getting the most out of esoteric hardware.
For all the mountains of posts about how the PS3, especially Cell, are not only 2x faster BUT architectural superior to the 360 CPU (which is a dog) so little has been done with it in terms of the industry. It doesn't mean it is useless, but the bottom line is developing hardware that caters to the market demands.
That said I do think we may see some compromises where there will remain a small number of very efficient, fast, serial oriented CPUs (like x86 OOOe processors) and a consolidation of the "FLOP" resources with many, very simple, cores ala GPUs. Performance per mm^2 is very high for GPU cores so if simple, peak FLOPs what you are going for that is the direction you would want to go.
A 2nd or 3rd generation Llano style CPU (a handful very fast OOOe CPU cores--meets the needs for the serial gameloop, "deadline non-efficient code," indie devs, etc) with the vector built on (could use extensions, on the same die similar to old-style ondie FPUs; giving you your high peak FLOP performance on die as well as be a setup and/or post processing monster for the GPU as well as physics and such if the libraries ever catch up) and then a normal GPU. Down the road 5-7 years after this style of system we could see single chip solutions with a fast OOOe core(s) on a sea of GPU styled cores.
Anyhow, after reading all the PS3 owners shrug over losing Linux because it was piss slow at basic tasks (like web browsing), I am not sure how argueing going for even simpler, more basic, cores than the PPE in the Cell processor and how serial performance isn't important ... my 1.4GHz Core Solo netbook runs FF faster
While ND and DICE may not need a faster main processor I think we have heard many developers note that speeding up their core loop and having some "forgiveness" for some bad code when crunch hits could really make a difference in a lot of games. Just because the industry is moving toward one direction doesn't mean it should be done overnight--probably the biggest problem with CELL. Sony tried to use their market share to force the industry in a particular direction. But they didn't anticipate the strength of the competition, the importance of tool chains, were late (and half baked), and the industry didn't buy into their vision. Right direction, but wrong road it seems.
