Nvidia Grace

2021

Nvidia CEO Jensen Huang shares with TAdviser plans for CPU market

In April 2021, Nvidia introduced its first in-house CPU processor, called Grace. It is based on the ARM architecture and is intended for use in the data center segment. Wide availability of the processor is expected in early 2023. Thus, the Nvidia product line for data centers will feature three types of processors: CPU, GPU (Graphic Processing Unit, graphics processor) and DPU (Data Processing Unit, coprocessor for data processing). Nvidia presented a roadmap for the development of each of them until 2025.

The development of Nvidia's own CPU processor based on ARM is an important event for the market, but hardly absolutely unexpected. In 2020, the company announced its intentions to acquire an ARM developer for $40 billion. As of April 2021, Nvidia is awaiting regulatory approval for this deal.

Jensen Huang believes that Nvidia can accelerate the process of introducing processors with ARM architecture in those markets where it works "(photo - datacenterknowledge.com)"

It is noteworthy that during the day after the announcements of Nvidia, its shares rose in price by 4%, while Intel and AMD at the same time shares fell in price by about 4% to^[1]. But can we say that there are prerequisites that the entry of Nvidia with its own processor into the CPU market for data centers dominated by Intel, on the tail of which AMD is trying to come, will entail its serious redistribution?

At the time of the announcement of Grace, Nvidia itself speaks of it as a niche product: it is primarily designed for the most complex calculations, including natural language processing, the creation of recommendation systems and the construction of supercomputer centers for AI. All these tasks are related to the analysis of huge data arrays that require ultra-fast calculations and a large amount of memory.

And, according to developers, to solve the "heavy" problems of the Grace-based system in conjunction with the Nvidia GPU platforms, they will be able to provide 10 times higher performance than the latest Nvidia DGX servers and workstations using CPU with the x86 architecture.

In a conversation with TAdviser during a call with the press, Nvidia CEO Jensen Huang made it clear that with the development of its own CPU, the company does not seek widespread competition in the CPU segment with Intel and AMD and plans to focus on the niche for which the Grace processor is originally designed. And it was created for large systems focused on working with a new form of software - data-driven software (approx. TAdviser).

According to Huang, the philosophy of Nvidia involves the creation of solutions that did not previously exist to solve problems that were not before, and Grace fits into this philosophy.

We are creating Grace to work with a new kind of applications related to understanding natural speech, AI models, recommendation systems that handle hundreds of petabytes of data, as well as for data-driven scientific calculations, "said Nvidia CEO in a conversation with TAdviser. - There are more and more such applications, this is a new segment. We expect to see a big leap here. Thus, Grace will be very useful for the industry.

Otherwise, Nvidia has "great partnerships" with Intel and AMD, he added. The company works closely with them in the industries of PCs, data centers, hyperscaling, supercomputer computing, as well as with new partners. For example, with Ampere Computing, which develops CPU, or with Marvell, which has developments in the field of peripheral computing and 5G systems. Nvidia also has a partnership with MediaTek, a major manufacturer of systems on a chip.

These are companies that have really great products, and our strategy, philosophy is to support them by connecting our platforms, for example, Nvidia AI, or Nvidia RTX, or Omniverse with their CPU. Thus, we can expand the market, "Huang concluded in a conversation with TAdviser.

Thus, Nvidia emphasizes that it will focus on a certain niche. However, this niche is not small. So, for example, analysts at MarketsandMarkets in 2019 predicted that the AI infrastructure market will reach about $50.6 billion by 2025, and called processors the largest segment of hardware solutions in this market^[2]McKinsey analysts in 2018 expected that revenue from AI-related semiconductors by 2025 could reach $67 billion^[3].

Along with the news about the development of its own CPU, it became known about the first incarnations of Nvidia's competition in this segment with Intel and AMD in the form of specific projects: the company announced plans to create a supercomputers Grace processor in 2023 at the Swiss National Supercomputer Center (CSCS) and. Los Alamos National Laboratory USA

In Switzerland, in particular, a new supercomputer called Alps will replace the existing one that uses Intel processors. It will be built by Hewlett Packard Enterprise. Her press release says the new supercomputer will use its HPE Cray EX systems. In the standard configuration, they contain AMD processors, but the latter is not declared among Alps partners.

A press release about the US Los Alamos National Laboratory on the Hewlett Packard Enterprise website states that it is also planned to use HPE Cray EX for a new supercomputer. The supercomputer already existing in the laboratory uses HPE Cray EX with AMD CPU processors. Nvidia clarified to TAdviser that in both cases, HPE Cray EX will use their new CPU. Thus, we can talk about the first large supercomputers using Nvidia processors without using Intel and AMD processors.

Meanwhile, in a cross-cutting fashion, Intel is preparing to compete with Nvidia in the GPU market: its first Ponte Vecchio GPU is expected to be released in 2021. By the end of 2021, the Aurora supercomputer for the Argonne National Laboratory should be created on its basis.

Announcement of the first ARM processor from Nvidia for data centers

In mid-April 2021, the American technology company Nvidia introduced the first in its history - ARMprocessor for, data centers which was called Grace. It is designed for the next generation, for the supercomputers first time it will be used in the Alps supercomputer at the Swiss National Computing Center (CSCS), as well as the Los Alamos National Laboratory. USA

The Grace ARM processor is named after the American scientist Grace Hopper, who once created the world's first compiler for a computer programming language. The development of Grace took more than 10 thousand man-years of engineering work. The high performance of the ARM processor is based on the fourth generation Nvidia NVLink interconnect technology, which provides a connection speed between Grace and Nvidia graphics processors at 900 GB/s. Grace supports the Nvidia HPC software development kit and the full set of CUDA and CUDA-X libraries.

Nvidia unveils its first Grace ARM processor for data centers

Grace is designed to solve the most difficult problems, in particular, related to the use of artificial intelligence technologies that control automated and robotic plants, as well as smart cities with hundreds of thousands of sensors, sensors and other IoT devices. Developers claim that the novelty is ten times higher than the capabilities of the fastest existing servers. Grace uses the innovative technologies of the LPDDR5x memory subsystem, which provides twice the throughput and ten times the energy efficiency (compared to DDR4).

The new architecture, according to developers, also allows you to achieve unified consistency of cache memory with a single address memory space, combining system memory and HBM memory of the graphics processor, which simplifies the programming process.^[4]

Notes