If nvidia worked with AMD to make a motherboard with a CUDA GFX integrated with the board and not use the PCI-E interface but a new one where it could actually take advantage of AMD's hyper-transport high-bandwidth to the CPU, then the CPU would no longer be (as much of) a bottleneck and then you could have perfect application execution with low latencies and high performance... But nooooooooo, they want to make money and don't want to share with others
Just get a good cpu and youre all set ;D
well, for gaming that would be okay, but this is more towards general applications such as physic calculations, stuff that you could use CUDA on, it would help in gaming a little, in the fact that you could use resources more efficiently and would therefore need a less power CPU/GFX
EDIT: they are doing something similar in new CPU's by putting it directly on the same die, bulldozer is one of them I think, and Intel is doing it with intel GFX, BUT nobody is doing it with Nvidia and CUDA.
So you mean ON-BOARD Nvidia graphics?
I'm sure there are more reasonable ways to reduce latency. Especially because onboard graphics make upgrading expensive.
EDIT
What I mean by that is, if your high performance physics processor, by Nvidia, was on the motherboard and you wanted to upgrade that, while keeping the low latency, you'd be inclined to buy a new motherboard. If you didn't care about latency, then you'd likely not have bought the board in the first place.
ruffleDK-this is different, the integrated graphics cards go through a logical PCI-E bus, limiting the speed and bandwidth to, this has direct access to the CPU, the difference is like when a cpu accesses a pagefile from an SSD, sure its fast, but the L2 cache of the CPU would be faster
LOLUMAD273-you could still upgrade, the integrated one would handle low latency-high priority stuff, while a add in card could handle lower priority things, stuff not too dependent on latency, also you could upgrade as long as you could add and remove the chips like a CPU, since you would only need to upgrade the GPU, the memory would be fine to leave alone, sure over the years they will develop better memory but they do the same thing with cpus, the make newer boards with newer sockets so your going to be upgrading your motherboard anyway, they will put the better memory on the newer ones, might as well do the whole package
EDIT: in fact you could upgrade the VRAM just like RAM, or they could use the same memory source, as long as the ram had a high enough speed and bandwidth