We haven’t talked about Nuvia (no relation to Nvidia) to-date, but I’ve had an eye on the startup since it came out of stealth mode late last year. Nuvia is a new CPU design company with an impressive industry pedigree, and it intends to challenge Intel and AMD for part of the overall server market, using its own custom ARM-based architecture.
That’s an eyebrow-raising claim, given how hard it has been for ARM-based companies to do just that. Ever since Calxeda imploded in 2013, we’ve seen multiple companies claiming they would bring ARM-based servers to market. Not all of those firms have been startups. In 2015, then-AMD CEO Rory Read confidently predicted ARM would have 15 percent of the server market by 2018. Qualcomm, Broadcom, and Applied Micro have all taken shots at the ARM server market, as has Cavium. Ampere is making its own effort in this arena. The market Nuvia is moving into is rather crowded.
The company, however, believes it has the pedigree and the business plan that will let it succeed, even thrive in this space. The company is led by Gerard Williams, president and CEO, who began his career at Intel before a stint with TI, followed by ARM. He led the CPU architecture teams on the Cortex-A8 and A15 before moving to Apple and designing one or more cores in the A7, A8, A9, A10, A11, and A12. Given that Apple has led the entire industry in terms of ARM performance, that’s no small pedigree.
NextPlatform has published an interview with John Carvill, VP of marketing, discussing what Nuvia hopes to accomplish and how it wants to accomplish it. Other prominent engineers working at the company include Manu Gulati, who designed the FPU on the original K7 and built HyperTransport and the northbridge design for K8. The K7 FPU is a noteworthy feather, because it marked the first time in years that any company but Intel had built an x86 chip with a top-notch floating-point unit. Here’s how Paul Hsieh described it, in 1999:
Most of the features of the K7 are really just super beefed up features that exist in the K6 (and P6). But what differentiates it is its radically out of order floating point unit. Through a combination of 88 (!!) rename registers, with stack and dependency renaming on a fully superscalar FPU AMD has created, with the possible exception of the 21264, what is probably the most advanced architecture I’ve ever seen. It also definitely presents a significant performance level above both the K6 and P6 architectures, despite the claims of some skeptical high profile microprocessor reviewers.
History proved Hsieh right on that point. Strong FPU performance was a particular strength of AMD CPUs for a number of years thereafter. Later in his career, Gulati then headed up SoC architectural design for a number of Apple SoCs. Another member of the team that might be familiar to old ATI fans is John Bruno, VP of system engineering. Bruno led the design team for multiple ATI mobile GPUs and then for the first Trinity Fusion APUs, which combined Llano silicon with a GPU based on the Radeon HD 5000s VLIW5 GPU. After that, he spent six years at Apple as the system architect on multiple iPhone designs. Collectively, the team is heavy on engineering chops.
Nuvia isn’t just betting on its design chops, though — it believes a specific focus on the hyperscaler market can give it a further boost. Nuvia’s argument is that server CPU performance has stopped growing, partly because data centers are increasingly thermally constrained. Multiple companies have already stated, for example, that they cannot simply continue to dedicate more floor area to ramping AI performance, and that better algorithms and hardware are needed to continue to improve performance.
Here’s Carvill, via NP:
“This is a server-class CPU, with an SoC surrounding it, and it is designed to be the clear-cut winner on each of those categories – and in totality,” says Carvill, throwing down the gauntlet to all of the remaining CPU players, who each have their own ideas about how to take on Intel’s hegemony. “And we are not talking about the incremental performance improvements that we have come to expect over the past five years. We are talking about really meaningful, significant, double-digit performance improvements over what anyone has seen before. It will be designed for the hyperscale world – we are not going after everybody. We are not going after the entire enterprise, we are starting with the hyperscalers, and we are doing that very deliberately because that’s an area where you can take a lot of the legacy that you have had to support in the past and push that aside to some degree and design a processor for modern workloads from the ground up. What we are doing is custom, and we will not be using off the shelf, licensed cores. We are going to use an Arm ISA, but we are doing it as a clean sheet architecture from the ground up that is built for the hyperscaler world.”
Nuvia wants to build a custom CPU architecture, designed for hyperscalers, by some of the same folks who have built Apple’s most successful SoCs. The company’s goal is to build a custom, proprietary SoC with a leadership-class server design. It’s a tall order for a firm that wants to challenge the dominant x86 hegemony, but the performance of Apple’s CPU designs really does speak for itself, and while Nuvia doesn’t have Apple’s resources or entire chip design team, it’s got enough industry veterans to watch for the company’s next move with interest. We’ll be curious to see how things shape up from here.
- Linus Torvalds Claims ARM Won’t Win in the Server Space
- AMD’s Server Market Share Could More Than Double by the End of 2020
- Rumor: Intel Prepping Xeon Cascade Lake Refresh to Fight AMD Epyc