What's so bad about sub 3000Mhz memory? especially these days? a good, tight 2666-2933 can squeeze pretty much everything outta these chips.
I would sin a bit and say that for many users these days, this just wouldn't cut it even for "light" esports who need high FPS with low settings.
I really really wanna see a 95W chip that has 4C\8T and a much beefier 14-16 cluster GPU
How can you claim that 2933 "squeezes everything" out of these chips when they have not been released? Until we can test them, we have no way of knowing how much performance is left of the table. Even 2400 MHz didnt saturate their old carrizo chips, and those were hamstrung by abhorid memory controllers and abysmal CPUs. Performance was STILL held back. With 2400 MHz memory, their 512 core iGPU was slower then a 384 core DDR3 dGPU. Their new one has 704 cores. 2933MHz will not be close to saturating it.
4000 MHz might get close, as that would supply similar bandwidth to the 512 core 7750 GPU, and VEGA is more bandwidth efficient then GCN 1.0. But it still has to feed the system itself, so it may take more then that.
EDIT: I'm wrong, that still would not be enough. The 7750, a 512 core GPU, had 72GB of dedicated bandwidth. DDR4 memory, in dual channel, at 4266 MHz, would supply 71.2 GB/s, and that would also have to feed the system and a 704 core GPU. If VEGA scales with memory the same way as GCN, it is likely that the limit on performance will be whatever the memory controller can bear, with RAM speed never becoming the limting factor, rather it being how fast the controller will support.
Raven ridge really will need either more on board cache (like the 128MB L4 on crystal well which allowed intel's 580 GPU to move past memory bottlenecks) or a triple or quad channel memory controller, especially if the iGPU grows more with the second generation.
For CPU only, sub 3000 is just fine. But for iGPU usage, you need as much absolute bandwidth as you can get.
/[sarcasm]
Please explain?