Home / Business / Nvidia embraces her arm and announces her intention to speed up all processor architectures

Nvidia embraces her arm and announces her intention to speed up all processor architectures



While the Top500 list was announced today at SAI Frankfurt with an upgraded Arm Petascale supercomputer in the top third of the list, Nvidia has announced its intention to make Arm a full-fledged citizen Processing architectures that it supports, with x86 and Power. Seeing increased activity and momentum for alternatives to x86 processors, Nvidia announced its support for Arm, paving the way for a "new path for highly energy-efficient and AI-activated exascaux supercomputers".

Once upon a time, in 2011, Nvidia developed a project to develop a complete Arm processor capable of powering personal computers, workstations, servers and supercomputers. The Denver project, as it was called, did not materialize in its original scope, but Nvidia eventually created Arm + GPU chips (Tegra / Xaviar and Jetson), designed to the integrated worlds of mobile phones, robotics, wearable games and autonomous vehicles.

The company has also invested in the development of Arm + GPU software through PGI, which it owns, and OpenACC, the open source software project it helps to support, but from the journalist's point of view, these plans appeared to be on-back, off-again or perennial in limbo.

In fact, the questions we had asked the OpenACC team about Arm support a few days ago had been politely deflected just a day before we would get the pre-briefing that Nvidia was going to reengage to arm himself considerably. Nvidia is cautious and pulls the slots as it turns out logical when considering the trajectory of Arm-64. The transition from the mobile platform to the general purpose server CPU has been chaotic, with several false starts and failures before finding traction – starting in the HPC space. Now, Marvell's Cavium Arm server chips equip the world's first petascale Arm supercomputer from Sandia National Labs. The plan and execution of the Exascale arm in Japan is well advanced and Europe is committed to manufacturing a European flea chip that will power its next exascale class machines.

Seeing the enthusiasm and dynamism of Arm (Addison Snell, Market Observer HPC on the market, at Intersect360 Research, points out that there are more exarch projects than Intel on Arm), Nvidia decided it was time to fully embark. support.

Nvidia plans to make available to Arm's ecosystem its full range of AI and HPC software, which accelerates more than 600 HPC applications and all AI environments by the end of the year. "The stack includes all Nvidia CUDA-X AI and HPC libraries, GPU-accelerated AI infrastructures and software development tools such as OpenACC's PGI compilers and profilers," says Nvidia.

"We have been working on this for two years, we have ported our platform, we have compiled and tested it. We are almost there and for us, it was not really a question of whether, but when, "said Ian Buck, general manager and vice president of accelerated computing at Nvidia,

"By combining Arm and GPU, we can offer energy-efficient supercomputers by taking all the computational burden, whether it's simulating flops or AIs, and unloading those parts of the calculation for that the processor can focus on its performance, which is fast one-thread execution, "added Buck.

Today's launch benefited from strong support from the ecosystem and more than a dozen expert testimonials from Ampere Computing, Atos, CSC, HPE, Jülich Computing Center , Marvell, Mellanox Technologies and others (see below).

Partner Cray, who is investing in the development of the Arm compiler, has been supportive. "At Cray, our vision of the exascale era is made up of systems that integrate artificial intelligence and analysis with modeling and simulation, systems that often require and require a diversity of applications. processor architectures and systems designed for data intensive workloads required in the fields of science, engineering and digital transformation. "Said Peter Ungaro, President and CEO of Cray. "We are excited to partner with Nvidia to help realize this vision of our supercomputers by leveraging their CUDA and CUDA-X HPC and AI software for the Arm platform and integrating them tightly into our Cray management and programming environment. compilers, libraries, and tools) already enabled to support Arm processors on our XC and future Shasta supercomputers.

EuroHPC explores Arm's potential as the engine of its future exascale systems. The Roadmap to the European Processors Initiative, announced at the EuroHPC Summit in May, includes several generations of ARM-based processors based in Europe with acceleration as well.

"The European Processors Initiative aims to provide the European Union with its own high-end, low-power, versatile and accelerator solutions. EPI and SiPearl, its industrial hand, consider very positively the new possibilities offered by Nvidia. The combination between the EPI-based microprocessor and the Nvidia accelerator could perfectly suit the basic equipment of future European exascale modular supercomputers, "said Philippe Notton, general manager of EPI.

The Japanese research institute RIKEN, which articulates its exascale strategy on custom-designed Fujitsu Arm chips, is also supporting this effort. "We have been a pioneer in the use of Nvidia GPUs on large-scale supercomputers over the past decade, including the most powerful ABCI supercomputer in Japan," said Satoshi Matsuoka, Director of the RIKEN Center for Science. computer scientist (R-CCS) and professor at the Tokyo Institute in Tokyo. Technology "At Riken R-CCS, we are currently developing the exascale exascale Fascaku exascale supercomputer based on next-generation Arm and are excited to learn that Nvidia's GPU acceleration platform will soon be available for systems based on on Arm.


Source link