NVIDIA, the graphics chip maker, announced at the International Supercomputing Conference (ISC) in Frankfurt, Germany on the 17th that it will support the CPU of the Arm architecture and open up ultra-energy-saving artificial intelligence (AI) for the high-performance computing industry. A new development path for supercomputers with exascale computing power.
Based on this partnership, NVIDIA will provide its full suite of AI and High Performance Computing (HPC) software to the Arm Business Ecosystem at the end of 2019, which accelerates more than 600 HPC applications and artificial intelligence frameworks.
NVIDIA said that its current system single-chip products for portable games, autonomous vehicles, robots and embedded artificial intelligence computing are all based on the Arm architecture. Therefore, this collaboration, NVIDIA's software stack, which includes a variety of NVIDIACUDA-XAI and HPC libraries, GPU accelerated AI framework, and software development tools such as PGI compiler with OpenACC support and performance analysis tools, To strengthen the development of both sides on the super computer.
NVIDIA has long been known for its PC graphics chip suppliers, which make the game look more realistic. But now, researchers are also using graphics chips in data centers to speed up artificial intelligence computing, such as training computers to recognize images.
In addition, Arm of the Softbank Group of Japan provides basic processor technology for most mobile phone chips. But the Arm processor is different from Intel or IBM chips, and Arm itself does not produce chips.
Instead, it authorizes the underlying technology so that others can use it to make chips. Companies such as AmpereComputing, which is led by Intel's former president, have been working hard to introduce these Arm-based chips into the data center to compete with Intel, which currently dominates the data center chip.
NVIDIA said that from a technical point of view, the plan to build a supercomputer with Arm would be a "heavy burden." However, NVIDIA is now willing to come in and cooperate with Arm because European and Japanese researchers hope to use Arm's technology to develop chips for supercomputers. This actually gives them a third choice beyond IBM and Intel. . Because in the open architecture, they can have more control over the Arm architecture.
NVIDIA further pointed out that according to the latest Green500 rankings, 22 of the world's top 25 most power-efficient supercomputers use NVIDIA technology. The key to making this happen is that a supercomputer with NVIDIA GPU can handle heavy processing tasks with a more power-efficient parallel processing CUDAGPU.
In this way, NVIDIA and Mellanox network structure work together to optimize the processing power of the entire super-computing cluster. In addition, NVIDIA's SXM3D package and NVIDIANVLink interconnect technology can expand the node scale in an extremely dense manner and drive the development of high-performance computing in supercomputers.