Did you miss a session at the Data Summit? Watch On-Demand Here.
Early last year, Nvidia unveiled the Grace processor, the company’s inaugural datacenter ARM-based CPU for AI and high-performance computing applications. But at the time, Nvidia declined to reveal a number of key details about Grace, including the number of transistors in the chip and where it might be used. That changed today at the company’s March 2022 GTC event, where Nvidia took the wraps off of the Grace CPU Superchip, the first discrete chip in the Grace portfolio.
Nvidia’s Grace CPU Superchip packs 144 Arm cores into a single socket. Consisting of two CPU chips linked via Nvidia’s NVLink-C2C — a new chip-to-chip interconnect — it’s designed to complement the Grace Hopper Superchip, which pairs a Grace CPU with Nvidia’s newly announced Hopper architecture-based GPUs.
“A new type of data center has emerged — AI factories that process and refine mountains of data to produce intelligence,” Nvidia cofounder and CEO Jensen Huang said in a press release. “The Grace CPU Superchip offers the highest performance, memory bandwidth and Nvidia software platforms in one chip and will shine as the CPU of the world’s AI infrastructure.”
The Grace CPU Superchip is based on Arm’s Neoverse family. First introduced in 2018, Neoverse incorporates the technologies of the Arm ecosystem that span from the edge to the cloud, particularly those that focus on the datacenter.
The specific architecture is Arm’s Armv9, which made its debut in May 2021. The Grace CPU Superchip — which Nvidia claims consumes around 500 Watts of power — features a memory subsystem consisting of LPDDR5x memory that offers bandwidth speeds of up to 1 terabyte per second.
Paired with Nvidia’s ConnectX-7 NICs, the Grace CPU Superchip can be configured into servers as standalone CPU-only systems or as GPU-accelerated servers with one, two, four, or eight Hopper-based GPUs. Nvidia says that it’s working with customers in high-performance compute, supercomputing, and cloud on applications for the Grace CPU Superchip, which it expects will become available — along with the Grace Hopper Superchip — in the first half of 2023.
While the datacenter market is largely dominated by x86 processors made by Intel and AMD — Intel’s server market share accounts for an estimated over 90% — the power efficiency of Arm chips have made them an increasingly attractive alternative for enterprise customers. Amazon Web Services is running Arm-based chips like Graviton3 in its datacenters, while supercomputers like Fujitsu’s Fugaku are leveraging Arm processors in combination with GPUs.
Nvidia previously announced that Grace will first be installed in 2023 in supercomputers designed by Hewlett Packard Enterprise HPE for the Swiss National Supercomputing Centre and the U.S. Department of Energy’s Los Alamos National Laboratory. And in 2025, Nvidia plans to introduce the next generation of the Grace architecture, code-named Grace Next, ahead of a new GPU family.
As analyst Roger Entner noted in a 2021 piece for Fierce Electronics, several companies — including Marvell, Ampere, and Huawei — have tried to make an impact with Arm chips on the server market. Internally, Amazon, Microsoft, Baidu, Alibaba, Facebook and others have designed custom Arm-based chips tailored to their own datacenters and needs. But not all of these efforts have been successful. Huawei spent roughly $3 billion funding a development platform and ecosystem around its Arm server chips, while Qualcomm was forced to exit the space after attempting to do the same.
“The Grace chip is designed with a general Arm cores design available to any [Arm] licensee and not specially customized to work with Nvidia GPUs,” Entner noted. “If Nvidia is allowed to buy Arm for $40 billion, the combined company would have many reasons to build specialized CPUs and an integrated architecture that will be considerably more efficient than standard ARM cores, thereby giving the larger Nvidia a more substantial edge in the server market.”
Of course, Nvidia ultimately declined to pursue its planned acquisition of Arm, citing regulatory pressures. Still, along with its BlueField DPUs, which also tap Arm-based chips for processing, Nvidia’s recent moves show that it’s serious about making waves in a server processor market that could be worth $17.89 billion by 2026.
“Leading-edge AI and data science are pushing today’s computer architecture beyond its limits — processing unthinkable amounts of data,” Huang said last year, when unveiling the Grace portfolio. “Using licensed ARM [intellectual property[, Nvidia has designed Grace as a CPU specifically for giant-scale AI and HPC. Coupled with the GPU and DPU, Grace gives us the third foundational technology for computing and the ability to re-architect the datacenter to advance AI. Nvidia is now a three-chip company.”
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn More