If games used FP than AVX wouldn't have been a dud for gamers.
The idea that a chip with 2006 IPC clocked at 1.6GHz with eight cores getting probably 6.7 scaling doesn't need AVX2 seems more like posturing and defending than anything else.
Just because a CPU has a new instruction set doesn't mean the software will instantly use it. Especially games are lagging behind on instruction sets nowadays because they typically need years from development start to finished product.
Also, AVX1 is focused on SIMD and high precision as well, I never expected it to be a big improvement for game coders. It's just that AVX2 is probably even less important for them.
Speeding up low precision instructions like PD did over BD with the divider could bring a lot more performance to the table for this usage scenario (but is also limited in how much further one can improve them).
While consoles are different than PCs, I was hoping we'd actually move past some of the more blatant limitations of consoles, but this processor is still going to cause considerable limitations over even current gaming PCs.
Consoles are typically SP, with small levels, and low levels of AI without a lot of them on screen. When they do go multiplayer they have a much smaller server size than PC's which are designed for lower end hardware not what is possible on current mid to high end PCs.
I don't see PS4 changing that without a large boost in integer ops which AVX2 would provide.
You have to put this in context first. The small levels were more a limitation of Ram than one of CPU power. The AI is very branch heavy afaik - in order low IPC Cell CPU? Yeah.
And the limited Multiplayer can easily bump into different bottlenecks. Put all Players in one screen? Possible GPU/Vram bottleneck. Let them drop a lot of mines across your field of view? Possible Ram bottleneck. Let them all go bonkers behind you? Possible CPU bottleneck.
Now look at what the next Generation will provide: A lot more GPU power per Pixel (assuming 720p vs. 1080p) and sixteen times more Ram and Vram.
Stronger singlethreads (more stable IPC thanks to better Branch prediction and OoO), more threads, more and faster cache. Especially calculation ticks of scripts and AIs should have a much more predictable length. That is already a massive boost without any new instruction set.