![]() ![]() These days, the GPU is getting a break, I feel. Games all need the GPU, and more of it over time than was ever asked of the FPU. The GPU has a longer run than the FPU, but that's because the software environment demands so much more. Floating point, now with advanced instruction sets, is part of the CPU and expected to be. It lived on through the 386 and 486(SX) and then became a block on the CPU. IIRC the FPU (floating point unit) became a "thing" with the 80287, a separate chip. I know it is not the same don't stab me with facts.īut the whole GPU-CPU thing reminds me of the days when there was no FPU on the CPU. For a CPU, this is its 元 cache, and we don't yet have gigabyte-scale 元 caches. You have no choice but to store everything in the fastest RAM you can find. Cray-YMP era), performing simple operations on vast amounts of randomly accessed and continually changing data, so caches don't help, the system has virtually no locality of reference. The immediate mode 3D rendering pipeline actually looks quite a lot like old supercomputers (e.g. Give AMD or Intel a motherboard with 2 GB of 100 GB/s memory onboard and they'll give you a CPU with an IGP that spanks mainstream GPUs. This means it has to be on the same PCB as the processor (GPU) using it. It can't even be on a shared bus, it has to be directly wired to the controller to keep the signal clean. It can't be on a DIMM, as the stub would kill the bus. The reason we need a video card is somewhere to put ferociously high bandwidth RAM. The GPU itself isn't the reason, we can do anything and everything the GPU can do on the CPU die. Click to expand.The whole reason we need video cards is that RAM tech has not outpaced CPUs and likely never will.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |