NVIDIA Brings CUDA to Arm for Energy-Efficient Exascale Supercomputers
Working With Arm to Provide an Open Architecture for Supercomputing
Today at the International Supercomputing Conference (ISC) in Frankfurt, NVIDIA announced that it is introducing support for Arm processors, allowing energy-efficient Arm-based systems to leverage NVIDIA’s GPU hardware for accelerated workloads. NVIDIA said it plans to make its full stack of AI and HPC software available to the Arm ecosystem by the end of the year, including all CUDA-X AI and HPC libraries, GPU-accelerated AI frameworks, and software tools such as PGI compilers with OpenACC support and profilers.
In the company’s press release, NVIDIA CEO Jensen Huang noted that once Arm support is enabled, all major CPU architectures — x86, POWER, and Arm — will be compatible with NVIDIA acceleration:
Supercomputers are the essential instruments of scientific discovery, and achieving exascale supercomputing will dramatically expand the frontier of human knowledge. As traditional compute scaling ends, power will limit all supercomputers. The combination of NVIDIA’s CUDA-accelerated computing and Arm’s energy-efficient CPU architecture will give the HPC community a boost to exascale.
Following the announcement last month that AMD would power what is expected to be the world’s fastest supercomputer at the Oak Ridge National Laboratory, NVIDIA is targeting energy efficiency rather than, for now at least, total performance. Alongside its announcement, the company touted the simultaneous release of a Green500 survey showing that 22 of the world’s 25 “most energy-efficient supercomputers” are powered by NVIDIA.
To be clear, this isn’t NVIDIA’s first involvement with Arm. For example, NVIDIA’s AGX platform includes the use of Arm-based processors and NVIDIA’s Deep Learning Accelerator (NVDLA) supports Arm’s Project Trillium machine learning platform. But it’s a major step in opening up broad support for NVIDIA’s products with hardware-accelerated workloads regardless of processor platform.
Ian Buck, VP of NVIDIA’s Tesla Data Center Business, explained that the company’s customers, particularly in Europe and Japan, prefer Arm over traditional options from IBM and Intel since it gives them more control over the performance and flexibility of their future supercomputers:
That openness ... makes it very attractive. What makes Arm interesting, and why we’re announcing support is, is its ability to provide an open architecture for supercomputing.
This week’s Arm announcement is the latest step from NVIDIA as it pushes further into the datacenter. In March, the company acquired Mellanox for approximately $6.9 billion. Israel-based Mellanox focuses on datacenter connectivity solutions such as high speed Ethernet and InfiniBand products and has a presence in many of the world’s top supercomputers.