What is the road ahead in GPU computing? Also, how does the GPU computing future look like?

Each vendor is following different strategy here. NVIDIA trying for less power, mixed precision, physically based rendering, stacked memory. AMD trying for CPU+GPU, low power.
As for me, the more future is towards CPU+GPU computing. So far it’s just GPU, now with number of cores doubling every 2 to 4 years, and GPUs having their control logic applying on fewer cores, they seem converging. Also, when the first processor invented, all the networking, FPU, memory, everything was off chip. Now most of them are on-chip. With HBM in hand, CPUs may also use them. And, GPUs already on chip, there will be a major OS/driver management update making GPU a part of CPU with out looking it as a separate processor.

Leave a Reply