The reveal of NVIDIA’s first CPU based on the Arm neoverse core sent waves through the semiconductor industry, as the company indicated its plans to enter the data center CPU market following the Arm acquisition (pending). Codenamed Grace, one of the main advantages of Team Green’s CPUs is a much higher CPU-to-GPU bandwidth (via NVLink) and nearly 30x times higher GPU-to-memory bandwidth, thanks to a network of closely knit CPUs and GPUs. This design is somewhat similar to AMD’s Infinity Fabric implementation in the Instinct MI200-Epyc Trento combination, although using a less traditional approach.
However, performance and scalability are only part of the equation. As we’ve seen with AMD’s Epyc lineup, despite featuring a much higher compute density (more cores), increased PCIe 4.0 lane count, and higher number of memory channels, the company is still only just scratching the surface after roughly 5 years of the initial introduction. This is primarily due to the tightly knit nature of the server and data center ecosystem.
Intel CEO, Pat Gelsinger didn’t seem very concerned when inquired about NVIDIA’s new Arm-based CPU aimed at AI workloads. He stated that the company is still the leader in the field of AI and this announcement by NVIDIA is an indication of the latter responding to the recent Ice Lake-SP launch which among others, features many new AVX512 and mixed-precision capabilities, aimed at improving the performance across data center workloads
Nvidia today announced some new chips for the data center market, with a particular focus on speeding up artificial intelligence applications with standard chips known as central processing units, or CPUs. The stock market reacted quickly. What’s your competitive position with them?
We announced our Ice Lake [a new microprocessor for servers] last week with an extraordinarily positive response. And in Ice Lake, we have extraordinary expansions in the A.I. capabilities. [Nvidia is] responding to us. It’s not us responding to them. Clearly, this idea of CPUs that are A.I.-enhanced is the domain where Intel is a dramatic leader.
We also have, with our Habana product line [a specialized A.I. chipmaker Intel bought in 2019], unquestionably laid out a very aggressive path, and our cloud partnership with Amazon is a great demonstration of that. So clearly, I’d say the idea of CPUs is Intel’s provenance. We’re now building A.I. into that and we expect this to be an area where we are on the offense, not the defense going forward.Intel CEO, Pat Gelsinger
Most analysts maintained a similar view, with Christopher Danely from Citi explaining that the Grace CPUs won’t gain much traction due to the “limitations of Arm-based processors.” He further states that the “lack of a high-quality software ecosystem” for Arm servers “will prevent them from being a threat.” Shortly after the GTC announcement, Intel’s stocks dropped by 5% while NVIDIA’s made some rapid gains.
Analysts from New Street Research shared a similar opinion. The analysts believe that NVIDIA’s announcement is more of a card bluff and instead of disrupting the market, will occupy a niche space. It’s more likely to be adopted by NVIDIA’s partners that are already using Telsa or V100/A100 GPUs in their data centers.