AMD, Intel & Nvidia Reveal Next-Gen Server Hardware at Supermicro Innovate

Nvidia

AI servers are big business, but they don’t come without some serious technical hurdles, as you require vast amounts of space, power, cooling, and a heap of money to build one that can compete in today’s market. However, with the work of Supermicro, ARM and NVIDIA, their new Grace MGX Servers are the next best thing to tackle all of these problems at once.

The Grace MGX Servers are named after computer programming pioneer Grace Hopper, who was one of the first programmers of the Harvard Mark I computer and was a pioneer of computer programming who invented the first compiler, to name but a few of her many achievements.

The collaboration between these three tech giants has allowed them to create the next generation of AI and HPC servers.

In the past, a system such as this would require a standalone GPU, such as the H100 Tensor core unit from NVIDIA, as well as a CPU and intermediary hardware. However, the Grace-based ARM CPU and Hopper-based GPU come as a single package, making them more compact, and easier to deploy.

With these AI and High-Performance Computing solutions now in a much smaller form factor, more hardware can occupy the same space and distributed cooling to allow for exponential gains.

These new systems are extremely powerful, featuring 144 ARM Neoverse V2 cores, and soldered on-module LPDDR5X memory, their all-in-one design means that the CPU and GPU can relay information between each other in a much faster way, and with systems like Chat GPT, latency and response times are king, so the reduction in bottlenecks is certainly a welcome one.