MLex-logo-colour

Comment: Nvidia’s Grace chip provides argument for regulatory approval of Arm deal

by Flavia Fortes

April 23, 2021

Nvidia’s move to compete head-to-head with Intel on high-end data center processors could provide a boost to the company’s rationale for its proposed acquisition of Arm. Nvidia is hoping that its new Grace chip, based on Arm architecture, is a step toward competing against Intel’s x86 processor in data centers. The acquisition of Arm, Nvidia can argue, puts it in a position to challenge the semiconductor behemoth.

Nvidia’s move to compete head-to-head with Intel on high-end data center processors could provide a boost to the company’s rationale for its proposed acquisition of Arm.

Nvidia is hoping that its new Grace chip, based on Arm architecture, is a step toward competing against Intel’s x86 processor in data centers.

The $40 billion acquisition of Arm, Nvidia can argue, puts it in a position to challenge the semiconductor behemoth.

Grace, named after computer pioneer Grace Hopper, is a central processing unit (CPU) designed for sophisticated data centers powered with artificial intelligence, and thus serves only a small slice of the market. It is, however, a first taste of what the Santa Clara company can do once it owns Arm.

Nvidia can argue that only after being fully integrated with Arm would it be able to create a general-purpose CPU for data centers to compete with Intel’s x86.

In a website to explain the transaction, Nvidia says Intel’s x86 has a near monopoly on servers that power AI. It says that Intel and AMD together have 97.4 percent of that market, with AMD holding less than 2 percent of that market.

Nvidia says that the dominance of x86 in the datacenter space has led to slower pace of innovation, limiting the potential of the entire industry; leading to lack of price competition; and constraining the ecosystem for Arm and the opportunities available to all its licensees.

Arm has been making its way into data centers, with companies such as Amazon designing chips for servers with it, but Nvidia says that “Arm has barely scratched the surface of what’s possible in datacenters and high-performance computing, which are critical to the future of AI.”

“Those markets have long been dominated by the x86 CPU architecture,” it added on its website discussing the deal.

The Grace chip competes with Intel’s central processors for high-end data centers, but Nvidia also wants to compete in the general purpose CPU market for data centers. But the question many ask is “why does Nvidia need to buy Arm to compete when it can just buy its licenses?”

— Ecosystem —

Arm, a British semiconductor and software design company owned by Softbank that serves about 90 percent of the mobile market, designs software for chips and licenses its intellectual property to customers.

In presentations justifying the deal, Nvidia says a general purpose Arm-based CPU that would compete with Intel’s x86 processor requires an ecosystem supporting Arm in the data center. That’s why several companies, including Nvidia itself, tried but failed to sell CPUs in the past.

“Over the past decade, several companies have developed general-purpose Arm-based datacenter CPUs, intending to compete with x86. But they haven’t succeeded in gaining traction or spurring a broad Arm ecosystem,” Nvidia says.

An ecosystem, in this case, is a network of software, developers and other peripherals that are developed to be compatible with a processor. The majority of software today is developed to work on x86.

Custom applications of Arm, such as Nvidia’s Grace, don’t rely on an ecosystem because they include a suite of software for limited applications.

An Arm-based processor developed by Apple was launched last year after Apple announced it would abandon Intel processors for its own M1 chip. Just like the M1, Grace is proprietary to Nvidia. It doesn’t benefit the ecosystem.

Experts don’t believe the Grace chip will compete with general purpose x86 CPUs because Nvidia’s chips were created for extremely advanced applications that require enormous data sets and represent a narrow section of the market. It has very specific capabilities and it’s not made for normal data centers, it’s understood

By buying Arm, Nvidia would be able to develop the ecosystem, infusing the Arm intellectual property with Nvidia’s knowledge and expertise. It is in its interest to make the Arm IP a world-class IP for CPUs, and Nvidia has promised to continue licensing Arm’s IP to all of Arm’s licensees.

Nvidia says developing this ecosystem justifies its $40 billion investment in Arm to create “the world’s premier computing company for the age of AI.”

Nvidia says its vision for the Arm deal is to grow Arm as a viable alternative to x86 for the general purpose data center market, which today is dominated by Intel. Intel has about 95 percent of that market.

— Critics —

Critics, however, are concerned that even though Nvidia may license its IP to competitors, there are other ways to disadvantage companies besides foreclosing access to the technology (see previous comment here).

For example, Nvidia could leverage Arm’s dominance in the mobile market by integrating its software and programming interfaces into Arm’s designs, thereby increasing the likelihood other vendors will make Nvidia’s chips and technologies the standard.

Nvidia currently licenses CPU technology from Arm for its own CPUs. After the deal, it would be able to sell its CPUs at higher margins because it would no longer have to pay royalties to Arm — another benefit of the deal that would give the company an advantage over its rivals.

Companies are also concerned that Arm’s model, with its licensing system and the flexibility of the architecture, could be affected with the deal.

But the argument that the deal could harm competition by creating a data center platform may have been weakened. Nvidia, which has strong positions in the data center market in graphics processing units (GPUs) and data processing units (DPUs), could complete the platform with a strong position in CPUs. However, with Grace, Nvidia has demonstrated its ability to enter the data center CPU market even without owning Arm.

But critics of the deal believe that evolving technology will result in information being processed interchangeably among the three types of processors, and the combination of the companies would create a dominant player across the board.