5 ms·
gpu/cpu unification has to happen some time in the next 10 years (in the 2010 thread i argue against it being next year, but it's inevitable at some point). th
by andrewcooke 17y ago
gpu/cpu unification has to happen some time in the next 10 years (in the 2010 thread i argue against it being next year, but it's inevitable at some point). the next amd architecture is a step in that direction.
edit: also, please, please, ultralight laptops with e-ink screens.
edit2: to clarify the above, cpus are currently tending towards more cores, while gpus are tending towards larger caches. both are trying to extend their area of application into tasks performed by the other, in the hope of more speed over a wider range of applications.
- DanielBMarkham 17y agoThink that will lead to real-time high performance ray-tracing in applications and games? Because that seems to me like the next logical step for the marriage. It's a problem domain where programmers are already struggling and it lends itself well to highly-parallel processing.
- andrewcooke 17y agoi'm not sure what it will lead to, to be honest. my best guess is that there's some new kind of interface (3d?) that will exploit the extra power.
- JulianMorrison 17y agoWhy would they unify? They are doing very different jobs (serial versus vector processing).
- jacquesm 17y agoBecause 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.
- JulianMorrison 17y agoIt 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.
- jacquesm 17y agoI'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.
- JulianMorrison 17y agoPoint, 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.
- andrewcooke 17y agoi'm not sure they're as different as you are making out. why couldn't you have a chip that's somewhere between a cpu, a gpu and an asic - it rewires itself to make a trade-off between cache+prediction v many alus+parallelism? maybe the asic part is a bit extreme, but fermi (the next gen nvidia) will have on-chip memory that is switchable between explicit local shared memory (gpu) and implicit local cache (cpu). why couldn't the same kind of flexible approach be made to instructions - longer+fewer v more+shorter pipelines?
- jacquesm 17y ago