This 2nm chip challenges the GPU

Latest update time：2024-07-22

Reads：

????If you hope to meet more often, please mark the star ?????? and add it to your collection~

Source: The content is compiled from hpcwire by Semiconductor Industry Observer (ID: ic bank), thank you.

Fujitsu and ARM are relying on open source software to bring energy savings to air-cooled supercomputing chips that will ship in 2027.

The Monaka chip will be manufactured using a 2nm process and is based on the ARMv9-a architecture. Two sockets will contain 144 cores each. The chip is the successor to Fujitsu's A64fx, which was used in the supercomputer Fugaku, which is the fourth fastest supercomputer in the world.

Fujitsu and ARM are adopting an open source tool chain, including development tools, libraries and compilers, to run on Monaka, which could have a spillover effect if the chip is used in Japan's next major supercomputer.

Fujitsu aims to use the chip for AI and HPC systems. The company is committed to reducing power consumption regardless of the workload, and ARM chips can achieve this.

“Data centers are critical infrastructure that supports our AI ecosystem, so the industry needs to build a new approach that enhances data center efficiency by fusing an energy-efficient hardware and software ecosystem,” Fujitsu said in a blog post.

The chip could be used in FugakuNEXT, the supercomputer that is expected to replace Fugaku. The new supercomputer is scheduled to be ready by 2030, but RIKEN's final decision will only be known after the feasibility study it is currently conducting is completed.

Due to the high energy efficiency of the system, Japan's RIKEN has decided to extend the operation of the existing Fugaku Nuclear Power Plant by ten years.

Fugaku is the world's most energy-efficient supercomputer on the Green500 list. Japan has not changed its energy-first approach, with a goal of reducing data center energy consumption by 40% by 2030.

Fujitsu is working with ARM to bring machine learning workloads to ARM’s Scalable Vector Extension 2 (SVE2) feature, which provides inference capabilities. Software optimizations for the CPU will also result in better energy efficiency.

Fujitsu is also backing the UXL Foundation, which is developing an AI development framework to replace CUDA. UXL’s parallel programming framework is a fork of Intel’s OneAPI, which allows AI applications to run on a variety of hardware with minimal software changes.

UXL is being heavily promoted by competitors other than Nvidia, including AMD and ARM.

This is bad news for Nvidia, which uses CUDA as the basis for its GPU sales. But Japan recently announced that it will purchase Nvidia's H200 GPUs for its upcoming AI supercomputer, ABCI 3.0. This supercomputer will provide 410 petaflops of double-precision scientific computing performance and 6 exaflops of AI performance.

Fujitsu argued that installing GPUs is expensive and consumes a lot of electricity, but Nvidia said its GPUs consume a lot of electricity but can calculate faster, thus improving research efficiency.

The Monaka chip will be equipped with 12-channel DDR5 and support PCI Express 6.0 and CXL 3.0. The air-cooled chip will adopt a 3D chip architecture.

RIKEN has replicated its Fugaku supercomputer in a virtualized environment on the Amazon Web Services cloud. Customers can run workloads similar to Fugaku on AWS.

The software framework for Aurora, the world's second-fastest supercomputer, is also based on the framework being developed by UXL.

Appendix: Introduction to Fujitsu Chips

Reference Links

https://www.hpcwire.com/2024/07/19/arm-fujitsu-targeting-open-source-software-for-power-efficiency-in-2-nm-chip/

Click here???? to follow us and lock in more original content

END

*Disclaimer: This article is originally written by the author. The content of the article is the author's personal opinion. Semiconductor Industry Observer reprints it only to convey a different point of view. It does not mean that Semiconductor Industry Observer agrees or supports this point of view. If you have any objections, please contact Semiconductor Industry Observer.

Today is the 3833rd content shared by "Semiconductor Industry Observer" for you, welcome to follow.