Nvidia Alliances Hinted at Arm Deal
In anticipation of closing its landmark deal to acquire chip IP vendor Arm Ltd., Nvidia is highlighting its partner ecosystem as it seeks to extend its considerable reach beyond personal computing and AI to HPC and the network edge. Not surprisingly, those platform alliances all use Arm-based architectures in their server processors while eyeing exascale and edge applications.
Speaking at this week’s Arm developer event, Nvidia CEO Jensen Huang laid out a strategy for preserving and integrating Arm’s business model as the GPU leader expands its roster of chip partners. In its efforts to create synergies between AI hardware and software, Huang noted partnerships with Fujitsu, Marvell Technologies and Ampere Computing, the Arm-based chip startup launched in 2018 by former Intel executive Renee James.
The trio serves as a “starting point” for combining the Arm platform with an array of CPU, GPU and new datacenter processors (DPU) to “complete a general computing platform,” Huang said.
Nvidia’s (NASDAQ: NVDA) collaboration with Fujitsu focuses on its Arm-based A64FX processor used in its Fugaku supercomputer, which currently tops the global supercomputing rankings. “It just kind of shows you the range of the Arm architecture from embedded controllers, which is tens of milliwatts, to the fastest supercomputer on the planet,” said Huang.
Last November, Marvell announced its ThunderX family of Arm-based server processors would support Nvidia GPUs. Marvell is also porting Nvidia Cuda-X AI tools, HPC libraries, AI frameworks and software development to its ThunderX platform.
The chip maker (NASDAQ: MRVL) said it would aim its GPU-accelerated 64-bit server processor based on the Arm v8-A architecture at HPC, cloud and edge computing.
Huang offered few additional details this week.
Nvidia’s Ampere A100 GPU released in May was named for the French physicist and mathematician. In its drive for additional synergies, Huang also announced further collaboration with the Arm-based chip startup with the same name.
Since its launch two years ago, Ampere Computing has steadily packed its cloud-native Altra processor with more Arm-based cores, announcing a total of 128 cores in June. It expects to begin sampling the chip by the end of 2020.
Duncan Poole, Nvidia’s director of platform alliances, noted at the time that Ampere Computing and Nvidia would collaborate on reference designs using Cuda-X tools to combine Arm-based CPUs and its GPU architecture. More recently, Nvidia partnered with Ampere Computing to extend its Mt. Jade server platform to cloud gaming. Its Altra-based systems includes two 80-core Arm-based SoCs, four Nvidia T4 GPUs and an Nvidia Mellanox BlueField-2 DPU.
Most of the chip partnerships were forged in the months prior to Nvidia’s $40 billion acquisition of Arm. Those alliances along with its deal for Mellanox provided clues to Huang’s overarching strategy of dominating AI hardware, software and development tools while challenging Intel’s server dominance in the datacenter and extending Nvidia’s reach into HPC.
The U.K. chip IP vendor was the missing piece: “Arm is finally at a point where it's going to evolve way past mobile devices,” Huang told the developer conference. “It's going to go into high performance, compute into the cloud, into the edge, into autonomous machines…. We ought to make the first cut, make the first commitment to bring our architecture to the platform.”
Arm CEO Simon Segars echoed Huang’s assertion that regulators would deem the acquisition pro-competitive and therefore beneficial for computing innovation. “AI computing is going to go to where the data is, and that's what I think we've got an even greater chance of delivering on,” added Segars.
--Editor's note: This story has been updated with additional details on Nvidia's collaboration with Ampere Computing.