Nvidia’s Arm-Based Grace Superchips Set To Hit Data Centers Next Year

Some say the new technology has a place in the market, but that older x86 technology from Intel is far-from-dead with superior support and software solutions available.

Nvidia on Tuesday announced the first deployments of its Arm-based Hopper GPUs and Grace CPU “superchips,” a move that could prove to be a game-changing compute shift in the data center market.

Asus, Foxconn Industrial Internet, Gigabyte, QCT, Supermicro, and Wiwynn will be among the first computer makers to release the new Nvidia chips, which will be used for a wide range of workloads spanning digital twins (which are digital prototypes created by artificial intelligence), AI, high performance computing, cloud graphics and gaming. Systems containing the Grace CPU Superchip and Grace Hopper Superchip are expected to begin shipping from those companies in the first half of 2023.

Paresh Kharya, senior director of product management for accelerated computing at Nvidia, said channel partners were “vital” to the success of the new chips. “We rely on our ecosystem to create products for the market, including servers… as well as higher levels of aggregation to the data centers,” he told CRN during a press briefing. “So channel partners are really critical to our business.”

Tom Morton, senior HPC account executive for Burnsville, Minn.-based Nor-Tech, said x86 will likely remain king of the CPU segment for the foreseeable future. “The Grace Superchip will be intriguing for large AI firms, hyperscale, and possibly the national laboratories,” he said. “Google, Amazon and Facebook will take a close look at Grace CPUs because they demand energy efficiency and performance per watt. But Intel and AMD control 90 percent of the server CPU market. It will be a matter of years before Grace CPUs begin to take a noticeable share of that overall market.”

The company is betting on Arm technology, at the very least, becoming a major factor in the data center business.

“A new type of data center is emerging – AI factories that process and refine mountains of data to produce intelligence – and Nvidia is working closely with our Taiwan partners to build the systems that enable this transformation,” Ian Buck, Nvidia’s vice president of hyperscale and HPC, said in a statement. “These new systems from our partners, powered by our Grace Superchips will bring the power of accelerated computing to new markets and industries globally.”

The new servers hitting the market are based on four new system designs featuring the Grace Superchips. Nvidia said the chips enable a broad range of compute-intensive workloads across a multitude of system architectures.

The Grace CPU Superchip features two CPU chips with up to 144 high-performance Arm V9 cores with scalable vector extensions and a 1TB-per-second memory subsystem. Nvidia claims the CPU provides the highest performance and twice the memory bandwidth and energy efficiency of today’s leading server processors to address the most demanding PHC, data analytics, digital twin, cloud gaming and hyperscale computing applications.

The Grace Hopper Superchip mashes the Hopper GPU with the Grace CPU in an integrated module designed to address HPC (high-powered computing) and giant-scale AI applications. The blazing speeds provided by Arm-based technology might be impressive, but some partners and observers say it’s too early to see the tech take over datacenters now dominated by x86 technology and Intel’s Xeon chips.

Roger Kay, a tech analyst with Endpoint Technologies Associates, says in a blog post that 80 percent of all simulation models run on x86 processors. Those processors have a wealth of technical support behind them. “So, it’s fair to ask why (Nvidia CEO Jensen) Huang is positioning Nvidia to be a major player in the digital twin space …while practically admitting that he doesn’t have the computing at the moment to back up the promise,” Kay wrote. “Unlike Intel, ARM can’t offer tens of thousands of engineers, a deep library of software, tools, and templates, and an ecosystem of partners to support cutting-edge application development. Even in the future, Nvidia may remain reliant on x86 processors.”

Kent Tibbils, vice president of marketing for Fremont, Calif.-based distributor ASI Corp., said Arm processing has interesting uses at the edge and back end, but x86 is still the dominant platform in the mainstream data center market. “I don’t see Arm moving into that bigger part of the market,” he said. “There’s potential for growth, but right now, that mainstream piece is dominated by Xeon and x86. But there are going to be all kinds of verticals within the market in general and different processes and technologies are going to find advantages – so there’s plenty of growth to be found for everyone.”

In an earlier interview with CRN, Intel’s Jason Kennedy, senior director for customer engagement and product management for Xeon, said Intel’s legacy data center platform has a “very strong foundation” and more than 75 million Xeon Scalable units have been sold since their introduction in 2017. “Flexibility and choice are important for those end users to be able to confidently bet on Intel base systems…”

But Nvidia sees things differently.

Nvidia’s Kharya said the company sees a $500 billion market opening in several areas, including the digital twin market (which is alone worth $150 billion). “This transformation requires us to reimagine the data center at every level, from hardware to software, from chips to infrastructure to systems.”

And it’s not just Nvidia that sees a big market potential for new data center solutions. AMD’s new EPYC processors are also taking aim at Intel’s data center stronghold.

“This interplay of interests is what they mean when they talk about ‘frenemies’ in Silicon Valley,” Kay quipped on his blog.