NVIDIA Accelerates Arm from Cloud to Edge

Estimated reading time: 4 minutes

Arm is at the heart of billions of devices — from mobile phones and autonomous vehicles to edge systems and the world’s fastest supercomputer, Fugaku, in Japan.

This broad range of systems is supported by some 13 million developers. NVIDIA is bringing this vibrant community an expanding set of tools of our own.

They include:

• NVIDIA AI – the industry standard for accelerating AI training and inference
• NVIDIA RAPIDS – a suite of software libraries to run data science and analytics on GPUs
• NVIDIA HPC SDK – compilers, libraries and software tools for high performance computing
• NVIDIA RTX – graphics drivers that deliver ray tracing and AI capabilities

For many years, NVIDIA has used Arm cores in its Jetson SoCs for edge computing and autonomous machines and more recently in networking products such as BlueField DPUs.

BlueField data processing units accelerate and secure networking and storage jobs for cloud, embedded and enterprise applications with support from partners and independent software vendors.

Bringing these resources to Arm, a leader in energy-efficient computing, increases customer choice and expands opportunities for Arm and NVIDIA partners across four broad sectors.

New Horizons in High Performance Computing
Arm is extending its processor technology with Scalable Vector Extensions (SVE) to meet the needs of the world’s largest and most powerful systems. NVIDIA engineers are already working to support SVE in Fujitsu’s A64FX, the processor that powers Fugaku.

SVE is the latest addition to the Arm Neoverse platform for data center computing and networking. It’s powering simulations that advance science on Fugaku, ranked No. 1 on the TOP500 list of the world’s most powerful computers.

Our work will help Arm partners and customers generate optimized software for systems that use a combination of Neoverse CPUs and NVIDIA GPUs. It’s the latest step in an HPC initiative announced in June 2019 to support NVIDIA’s CUDA software for accelerated computing and AI on Arm.

In addition to the NVIDIA HPC SDK, software for NVIDIA Mellanox InfiniBand, the networking backbone of HPC, is fully supported and deployed on Arm servers today. And engineers are expanding support for Arm in NVIDIA Magnum IO, software that maximizes storage and network performance for multinode systems.

Games in the Cloud Create Serious Opportunities
In cloud computing, NVIDIA is engaged with Arm server SoC and OEM partners on a range of fronts.

For example, NVIDIA partnered with Ampere Computing to extend its Mt. Jade server platform to cloud gaming. Ampere Altra-based systems pack two 80-core Arm-based SoCs, four NVIDIA T4 GPUs and an NVIDIA Mellanox BlueField-2 DPU. Together they can serve 128 gaming users simultaneously, streaming sessions for the Android-in-Cloud services popular with China’s growing number of 5G smartphone users.

In addition, NVIDIA engineers have been porting code to Arm and developing new tools to optimize how cloud games are encoded, rendered and streamed to and from Arm-based servers and users. These tools are widely available to Arm server OEMs such as GIGABYTE, Inspur and Wiwynn.

NVIDIA knows firsthand the potential of cloud gaming, thanks to its experience running its own service, GeForce NOW.

Beyond gaming, cloud service providers are embracing Arm-based servers for machine learning, storage and other applications, accelerated by GPUs. That’s why NVIDIA provides a range of GPU management and monitoring tools for Arm-based servers, including the NVIDIA Container Toolkit to run Docker containers on Arm with Kubernetes.

Bringing Enterprise AI to Every Company
The edge of the enterprise network is the next big frontier in computing. The NVIDIA EGX edge AI platform will support all major processor architectures, including Arm.

Recognizing the need for energy-efficient computing at the edge, NVIDIA made Arm a centerpiece of its platforms for accelerated computing and AI across vertical markets. These platforms typically pair a GPU with a multicore Arm-based CPU in embedded modules like Jetson.

We pair tailored versions of these modules with developers kits specific to sectors such as healthcare (Clara), robotics (Isaac), autonomous vehicles (DRIVE) and more. They provide a rich set of opportunities that Arm software partners and SoCs providers can leverage to enable enterprise AI.

And they’ll only get richer with NVIDIA’s intention, stated in its acquisition announcement, to make its IP available through Arm.

The Shape of Things to Come
Windows PCs and Chromebook from multiple providers already run on Arm processors with more consumer platforms on the way. That’s fueling ideas about the future of personal systems.

NVIDIA has a long history of supporting PCs of all shapes and sizes with its GPUs, including Arm-based systems. More than 70 percent of gaming PCs use an NVIDIA GPU, according to a recent survey.

The future holds opportunities for AI-accelerated personal systems, rich in graphics and connectivity, as powerful as today’s PCs and more energy efficient. An NVIDIA reference design currently used in healthcare markets is an example of capabilities the Arm ecosystem can leverage into markets such as personal systems.

The Tools that Build the Platforms
New platforms are made possible by advanced technologies like the Arm portfolio of processor IP that NVIDIA supports with a wealth of tools.

The CUDA toolkit for Arm brings to Arm’s ecosystem NVIDIA’s GPU-accelerated computing and AI capabilities. Through CUDA, developers get access to TensorRT for deep learning inference, DeepStream for video analytics and more.

NVIDIA’s offers a full suite of Nsight visual developer tools supporting Arm-based SoCs and servers with profilers, debuggers and more. You can access all our tools, SDKs and platforms through our developer site.

NVIDIA has been supporting Arm for more than a decade, and we’re just getting started. We’re excited to be part of an expanding community that powers everything from smart cards to supercomputers and so much more to come.

We have lots of ideas about what’s possible, and there’s much we hope to learn from the 13 million developers in the Arm ecosystem.