Is bandwidth all? I think it heavily depends. Like with the multi-die Xenos solution, or the multi-SPU Cell solution. It totally depends on what you do where, and what things have to be communicated. Bandwith by itself isn't saying anything at all.
And how about latency? Like, a CPU has really low latency in general, while a GPU has a high one? While the GPU reaches an order of magnitude more FLOPS? As long as you have lots of things running in parallel and can switch between them, latency isn't important. Sometimes it's even necessary to get things to execute in the right (dependent) sequence. And sometimes (like with CPUs, especially when they don't do out-of-order execution, or with GPUs, when their FIFOs are full), it just generates stalls.
But, how do you know? It depends. Not even the hardware layout by itself will tell you what effects are going to happen. Only when actually running a game will you know. And with some optimizing, it might not even be much of a problem after all, as long as there are enough other things left to do for the various processing units.
In short: it depends.
And how about latency? Like, a CPU has really low latency in general, while a GPU has a high one? While the GPU reaches an order of magnitude more FLOPS? As long as you have lots of things running in parallel and can switch between them, latency isn't important. Sometimes it's even necessary to get things to execute in the right (dependent) sequence. And sometimes (like with CPUs, especially when they don't do out-of-order execution, or with GPUs, when their FIFOs are full), it just generates stalls.
But, how do you know? It depends. Not even the hardware layout by itself will tell you what effects are going to happen. Only when actually running a game will you know. And with some optimizing, it might not even be much of a problem after all, as long as there are enough other things left to do for the various processing units.
In short: it depends.
