The Era Of General Acceleration Has Begun

In this new era, compute engines are going to proliferate, offering varying types and capacities of compute and memory to match a set of workloads that is diversifying faster than the SKU stacks of the major CPU vendors. It has been decades since just being able to manipulate integers was sufficient in a compute architecture, which saw the rise of outboard math coprocessors and then integrated vector engines and now, with the advent of artificial intelligence, the rise of matrix math units to accelerate AI inference and in some cases AI training, too. But all kinds of other functions that might otherwise be done by CPU cores are being done by accelerators, often integrated on cores, sometimes integrated in the CPU package and other times in outboard accelerators such as GPUs, FPGAs, or DPUs.

Powered by Intel

White Paper from