Comment by jacquesm
17 years ago
Because there are two converging trends at the moment:
CPU makers getting in to the GPU arena:
AMD bought ATI, intel has announced a CPU/GPU combo
NVidia is going in the direction of more and more general computation capability with their offerings (see the gt 300).
Sooner or later those trends will meet somewhere in the middle.
I predict this will happen as soon as NVIDIA offers a product that has enough general compute capability to run a linux port.
It will become a general co-processor, integrated on the low end, separate on the high end, never unified. The two types of chip fundamentally don't do the same job. They can't run each other's programs.
I've heard that before about floating point co-processors, they're definitely unified now.
Remember the Weitek and the 387 ?
In the end it's a cost-savings issue, as cpus get more cores they become more and more like GPUS, as GPUS acquire more general purpose instructions they become more like CPUs. Those lines will meet, once they're on one die the 'budget' can be used more efficiently by looking for ways to integrate them more tightly.
There is nothing inherently different about general purpose computations vs the kind of vector processing that a GPU is good at, at the end of the day it is all calculations, and more and more in parallel.
I fully expect that at some point even DRAM will be part of the CPU.
Point, but I'm not sure they're comparable. Both CPU and GPU are something you can never have enough of, because their task list expands to meet supply. Contrast: once you have enough FPU, you're done.
2 replies →
ah, ok. so change the word unification to integration - i am not being that specific.