Nvidia And Arm Are Aligned On The Future Of Arm, But Is Everyone Else?

Arm revealed its newest Neoverse CPU cores for data center and network processing applications. These new cores, the Neoverse V1 and Neoverse N2, offer significant improvements over the original Neoverse N1 core introduced two years ago and used in the Amazon Web Services’ (AWS) Graviton2 server processor. This news comes on the heels of Arm’s announcement of its next generation ArmV9 instruction set architecture. While the V1 will still be using the ArmV8 architecture, the Neoverse N2 core is the first announced core to support ArmV9. These announcements by Arm are all taking place while Nvidia is trying to acquire Arm from its present owner Softbank, which I covered in a previous Forbes article.

MORE FROM FORBESNvidia To Buy Arm – What’s Next?

Just two weeks earlier, Nvidia announced that it is building its own Arm CPU called Grace for high-performance computing and large-model training. Grace is further proof that Nvidia values the performance possible with the Arm architecture and Nvidia’s desire to push the Arm architecture to the highest performance platforms. Nvidia didn’t reveal the name of the Neoverse CPU core in Grace, but it is more than likely either a V1 or N2, or an improved version of one of the two cores. Of the two cores, the V1 is the highest performance core designed with scalable vector extensions (SVE) instructions for high-performance computing (HPC) applications. While the V1 is the highest performing core of the two, it is based on the ArmV 8.6 architecture, rather than the ArmV9 like the N2. It is possible Nvidia is using the Neoverse N2 for Grace as the CPU isn’t responsible for the compute heavy lifting reserved for the GPU accelerator.

Nvidia’s primary reason to buy Arm is to push the performance of the architecture higher to offer an alternative to the x86 architecture from AMD and Intel. With Arm moving to its 9th generation architecture with improved security features and SVE2 SIMD vector extensions, Nvidia can improve power efficiency and scalability of its platforms while advancing the Arm architecture in the data center.

In terms of specifications and simulated performance numbers, these new Neoverse cores go a long way toward closing the performance gap between Arm and x86. Arm’s projection on performance per core and per socket are based on the SPEC benchmarks, which have been used historically to measure and compare CPU performance. However, until there’s working silicon with the Arm V1 and N2, the delivered platform performance on real-world workloads is still difficult to judge.

Arm And Nvidia – Perfect Match Or Perfect Threat?

While there are alignment and synergies between Arm and Nvidia, there have been many critics of Nvidia’s acquisition plan for Arm. In a scenario where Nvidia is in charge of Arm’s roadmap, one would expect even higher performing CPU cores because of Nvidia’s focus on silicon and system performance. That should be a good thing for Arm licensees that care about performance, which would include data center, high-end smartphones, networking, telecommunication systems, and Arm-based PCs. That said, there is certainly concern from some Arm customers because Nvidia is a competitor to many of Arm’s licensees in key markets, such as automotive, data center, telecommunications, and robotics.

There’s an expectation that Nvidia would have a very “hands on” approach to Arm’s development plans and roadmaps and would try to accelerate Arm’s roadmap by investing in R&D resources, providing direct engineering guidance, as well as intermixing Nvidia intellectual property (IP) with Arm IP. By combining Arm CPUs with Nvidia GPU’s, Nvidia can also offer more choice for third-party SoC support to data center and cloud customers, in addition to AMD and Intel. The challenge for Nvidia will be finding the right balance between directing Arm toward their joint goals but doing so in a way that doesn’t damage the latter’s position in the marketplace as an independent provider of IP.

Arm’s independence is often described as an asset. But in fact, its status as a pure IP player — with no silicon products, platform solutions, or associated software business – can ultimately slow its pace of growth. For example, Arm has limitations on developing or benefitting from platform solutions. It makes investment decisions solely based on its pure IP licensing model. This business model puts limits on the level of investment Arm can make in new markets. The companies who have benefited the most from Arm are those companies with strong platform businesses.

In markets such as datacenter, where IP beyond CPU cores are requirements for test chips, reference boards, reference systems and software elements, Arm is at a disadvantage. As a result, Arm’s relative investment compared to the large x86 ecosystem players is relatively small. Arm’s business model doesn’t provide the scale of the x86 competition nor does it allow for system level innovations that require other parts of the platform. So, for years, x86 has maintained its dominant position, notwithstanding predictions that Arm was poised for a breakthrough. The success of Arm in this market is dependent on licensees making appropriate investments and developing system-level innovations.

Nvidia’s focus on end solutions (chips, boards, and systems) will complement Arm’s IP and accelerate its push in new areas, such as the datacenter, while benefitting competing solutions from other Arm partners. Nvidia says it will keep Arm open and license all Arm technology as soon as it is available, which is Nvidia’s only rational choice. Otherwise, it would destroy the value of the Arm business, right after paying $40 billion for it.

Alternatives To Nvidia

Under Softbank ownership Arm has been making large investments in growing Arm’s market position in IoT, automotive, and data center. But an independent Arm would have trouble funding such heavy long-term investments.

It’s clear that Softbank wants to sell the company and make some return on its investment. It could be argued that SoftBank overpaid for Arm, which was one reason why it was difficult for Arm shareholders to resist the offer. For their part, Nvidia was willing to pay $40 billion for Arm because it opens a new business opportunity through licensing Nvidia IP through Arm channels. Additionally, Nvidia’s investments will position the company to reap more of the billions of dollars associated with the server, PC and edge AI markets.

The alternatives to Nvidia are few. One would be to have a group of Arm partners acquire the company from SoftBank. This would be similar to the investments Apple, Intel, and other companies once made in Imagination Technologies to keep it as a 3rd-party IP provider. However, the competitive nature of the Arm ecosystem and the $40 billion price tag sought by SoftBank makes this scenario unlikely. Additionally, this would not necessarily help ARM address platform-based innovation. The other alternative is to take Arm public through an IPO, which was SoftBank’s original plan for Arm. This would still be feasible and might generate a significant sum given the financial market’s recent appetite for tech stocks. However, there is no guarantee that it would fetch the $40 billion SoftBank is seeking, nor would it provide additional investment back into Arm.

So, it comes back to Nvidia. The company is willing to spend the very hefty sum to acquire Arm and is committed to investing in Arm to push it to higher performance levels that will create a competitive alternative to x86. What concerns some Arm licensees is the unknown. If Nvidia takes control of Arm, does Nvidia get first preference on new Arm cores? Will Nvidia raise the price of Arm licenses and royalties to pay back the investment? How much independence will Arm have under Nvidia ownership? How open will Nvidia be to work with Arm licensees to decide on the architecture’s future? Nvidia has said it plans to keep Arm’s licensing model open to all and is incentivized to keep the Arm ecosystem vibrant and expanding. If that’s Nvidia’s plan, then Arm licensees can expect to get the same Arm IP (and more) on an accelerated track.

As far as some licensees are concerned, they would be happy to see SoftBank continue as a relatively benign owner. That would preserve the status quo of a minority of large architectural licensees who dominate their market segments and a majority of smaller players and startups. However, it’s uncertain if the status quo will enable Arm and its partners to compete against the giant x86 competitors in multiple markets. Then the real question becomes which are the best alternatives for Arm at this point?

I can’t address the issue of regulatory approval in the various countries interested in Arm’s business, because that is extremely hard to handicap. Nvidia themselves have set a long timeline for the process to play out, and that is Nvidia’s burden to bear. Then again, the question becomes if the regulatory agencies shoot down the deal, what would be the future of Arm? Arm IP is too important to the tech world and the global economy to be treated lightly. When you consider the alternatives for Arm’s growth and strengthening it for competition against the x86 players, being acquired by Nvidia may be the most viable option for Arm and its ecosystem.

The author and members of the TIRIAS Research staff do not hold equity positions in any of the companies mentioned. TIRIAS Research tracks and consults for companies throughout the electronics ecosystem from semiconductors to systems and sensors to the cloud. Members of the TIRIAS Research team are tracking all the developments in AI and enterprise computing technology and have consulted for Arm, Nvidia and other companies focused on PCs, Graphics, AI and enterprise computing solutions.