I feel like future CPU hardware is looking a little gloomy over the next few years specifically for the high-end market.
We’re in a transition period with hardware and it kinda sucks since the software is lagging behind the mass parallelism hardware that keeps getting expanded on.
There’s a lot of great potential in things like phi and their compiler work, but it doesn’t stop the fact that a lot of things don’t improve in speed with multithreading and can even slow down in some cases because of the distribution overhead. Though maybe compiler development will solve that.
Things like the mental ray software renderer making use of the GPU to output AO passes is neat, and it’ll be nice to see more and more things dumped onto PCI GPU’s or coprocessors. It’s also neat to see OpenSubdiv dump all its calculations onto the GPU.
I guess it remains to be seen if renderfarms full of PCI rack servers will render frames faster than dense dual-CPU blades or if they’ll be more cost efficient. Maybe renders will happen faster by adding more large GPU’s, but right now they’ll also go faster by adding another computer or another CPU socket.
I just wonder about if transitioning to mass parallelism will actually bring any real benefit. Similar to how a really high quality GPU render takes about as long as an identical-looking high quality CPU render.
