It is an exciting time to be covering semiconductors as a tech analyst, with more players on the scene than ever before, developing custom and general-purpose silicon for increasingly specific use cases. That silicon activity is a direct response to the explosion of the data processed by businesses and ultimately processed and delivered to consumers. Of course, you have your large, traditional silicon vendors—Intel, AMD, NVIDIA, Qualcomm, and the like. However, perhaps more interesting (and more fluid) is the jockeying of many startups that have emerged in recent years. These ranks include NUVIA, a Santa Clara, CA-based silicon design company, founded in early 2019 with a mission of developing “the world’s leading server processor.”
This week NUVIA announced an important milestone on its journey: an impressive $240 million close to its Series B funding round. Let us check in on NUVIA’s progress and how this fundraising haul will punt the ball further down the court for the ambitious upstart.
Who is interested in NUVIA’s mission?
A haul of that size is a clear indicator that NUVIA has piqued the curiosity (to put it lightly) of some deep-pocketed, influential benefactors. The funding round was led by Mithril Capital in partnership with Sehat Sutardja and Weili Dai (founders of Marvell Technology Group), funds and accounts managed by BlackRock, Fidelity Management & Research Company, LLC., and Temasek, with additional participation from Atlantic Bridge, Redline Capital, Capricorn Investment Group, Dell Technologies Capital, Mayfield, Nepenthe LLC, and WRVI Capital.
The $240 million from these investors adds to its Series A round of funding (read here), which raked in $53 million last November and included several of the same firms. Furthermore, NUVIA says that several of these could potentially invest further for a Series C round of funding, which would likely occur around the time of NUVIA’s first silicon. All of this to say, these types of firms do not just shell out money like that for nothing—clearly, they believe in NUVIA’s mission and product vision.
What is NUVIA building?
This brings us to what, exactly, NUVIA is working on behind its closed doors that have these investors so intrigued. Its founders, John Bruno, Manu Gulati, and Gerard Williams, are all ex-Apple engineering leaders seeking to challenge Intel and AMD’s x86 primacy in datacenter silicon. To do so, they are attempting to build an SoC with “category-redefining performance, energy efficiency, scalability, density, and TCO”—an SoC they’ve codenamed “Orion.”
Then, there’s the company’s “Phoenix” CPU currently in development. If the Orion SoC is NUVIA’s long-game, “Phoenix” is the building block that will make Orion possible. NUVIA touts Phoenix as a “completely new design of the CPU pipeline,” and a significant departure from Intel’s and AMD’s x86 and other Arm-based architectures for the datacenter. Typical metrics cloud providers use when comparing datacenter silicon is TDP or the performance per TCO (total cost of ownership) of the chip. Another thing these cloud providers must consider is how these chips’ TDP fits into its tightly controlled power budgets. NUVIA expects to beat Intel and AMD’s offerings on both higher peak performance and energy efficiency, achieving the balance that all high-performance data center CPUs aspire to fulfill.
To do so, NUVIA says Phoenix maximizes its memory bandwidth and core utilization. It also says it will not require extra cores to compensate for a single-thread performance deficit, nor will it need to employ boost clocks, which often inflate numbers for the sake of marketing. The problem with these boost clocks is that it is not always indicative of real-world, continuous, server performance, hence “burst.” For a more accurate accounting of power and efficiency, I believe server SoCs, particularly cloud, should be tested under the conditions they run in the real world—heavily loaded. Running it on less in an attempt to juice the numbers does not paint a full picture for customers. Ideally, according to NUVIA, an SoC should complete workloads in the shortest amount of time possible, while drawing the least amount of power possible. That is the goal of Phoenix.
Some surprises from NUVIA’s CEO, Gerard Williams
According to Gerard Williams, whom I had the chance to speak with over video last week, the money garnered will enable NUVIA to get its Orion silicon in hand by the end of 2021. There will be two different versions of its reference platform up and running, some 4-6 months later in 2022. Williams says it has made the technology selections for its reference boards—that is, all the components that go into them, such as packaging, sockets, DRAM, PCIe, thermals, and power delivery.
Lastly, and very interestingly, Williams said that the company’s GeekBench single-threaded benchmarks correlate to what I would consider higher-order server benchmarks like SPEC. Now isn’t that interesting? Next up, I could see the company releasing performance expectations for multi-threaded applications.
While this fundraising haul is impressive, NUVIA certainly still has its work cut out for it. A struggle all new startups face is finding its foothold without getting trampled by the elephants in the room (read: Intel and now AMD). I can surmise that AWS has had success with its custom, Arm-based SoC, Graviton2, but not every cloud provider can or should roll its own silicon. I believe that NUVIA has a leg up because its founders have proven resumes working on Apple’s silicon efforts, and there is a market need for dominant performance per TCO. With experienced leadership, a clear strategy, and now a boatload of funding to invest, NUVIA is in a good position. I will be following the rollout of Orion and Phoenix over the next few years with interest.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.