The Grace Superchip CPU, which was introduced in early 2022, has been revealed by NVIDIA CEO Jensen Huang to look exactly as it does in the renderings. The actual product’s release was postponed from the first half of 2023 to the second half of that year.
Two Grace CPUs, totaling 144 cores, are combined into the Grace Superchip. Grace is built using the more energy-efficient Arm architecture, which is optimized for workloads involving artificial intelligence. Because x86 chips are made for general-purpose computing, they are less suitable for running these kinds of workloads.
Grace is compact due to its low power consumption, and two units of the Grace Superchip (a total of four CPUs) can fit inside a typical 1U server. NVIDIA’s benchmark shows that it can handle microservice workloads up to twice as well as x86 servers at the same power level.
The Grace Superchip has already started to arrive as samples to server manufacturers, and it should go on sale in the second half of the year. ASUS, Atos, GIGABYTE, Hewlett Packard Enterprise, QCT, Supermicro, Wistron, and ZT Systems are the brands that have been officially confirmed.
Given that the Grace Superchip is created specifically to meet the demands of AI workloads, its introduction could represent a significant advancement in the field of artificial intelligence. For companies and organizations that heavily rely on AI, its energy efficiency and capacity to handle microservice workloads better than x86 servers may make it an appealing option. It might also be simpler to integrate into current server infrastructure due to its small size. It will be interesting to see how it functions in actual situations and how its performance and energy usage stack up against those of other CPUs.