Description
The rapid emergence of AI and its unique and often demanding computational requirements are having a significant impact on the overall design and development trajectory of advanced HPC component accelerators. Prior to the rise of AI starting in 2019, such accelerators, particularly graphic processing units or GPUs, were largely seen as a complementary, specialized compute resource to facilitate offloading from the system’s central processing unit (CPU) a narrow set of computationally-intensive operations integral to science and engineering workloads. With the emergence of AI, however, the role of accelerators, particularly GPUs at this time, have become more central, requiring significant changes in their design and in many cases becoming the defining compute engine for an increasingly large number of HPCs.
