NVIDIA Shows Product Roadmap for Grace CPUs and New Data Centers

NVIDIA CEO Jensen Huang loaded data center products with the newly announced NVIDIA Grace CPU on April 12, the first day of the NVIDIA GTC conference. Presented the map. Huang also announced partnerships with companies such as Amazon Web Services (AWS) and Ampere Computing, which combine Arm-based processors with the company’s GPUs as part of the company’s data center strategy.
In his keynote speech, Huang said Grace “provides the ability to rebuild all aspects of a third underlying technology for computing and a data center with artificial intelligence (AI) in mind.” It was. The Grace chip is designed for terabytes of accelerated computing (specialized high-speed processing) and will be used in data centers, similar to NVIDIA’s GPUs and DPUs (data processing units). Become. “The architecture of each chip has a two-year cycle, and performance improvements will be made during that time,” Huang said. “We will focus on the x86 platform for one year and the Arm platform for another year.” So every year you’ll see our amazing new product … three chips, annual enhancements, a single architecture. ” NVIDIA GPUs are already playing an important role in the data center and are growing the business rapidly. In February’s fourth-quarter earnings, data center sales were $ 1.9 billion, up 97% year-over-year. For the full year, it increased by 124% to $ 6.7 billion. In September, it announced that it would acquire Arm from SoftBank Group for $ 40 billion (about 4.2 trillion yen). “Arm is (the architecture of) the most popular CPU in the world, and for good reason,” Huang said in a keynote. Although Arm is widely used in the mobile and embedded markets, its adoption is “just beginning to grow” in areas such as the cloud, enterprise, and data centers at the edge. “We can accelerate the spread of Arm in the markets we serve.” (Huang) In partnership with AWS, Huang will offer NVIDIA GPUs on “Amazon EC2” instances based on “AWS Graviton 2”. Announced that it will go. AWS has been offering Arm-based CPUs since late 2019. “This partnership will bring Arm to the most sought-after cloud workloads: AI and gaming in the cloud,” Huang said. EC2 instances allow game developers to run “Android” games natively on AWS. Regarding the partnership with Ampere Computing, NVIDIA also announced the NVIDIA Arm HPC Developer Kit, a developer kit for science and cloud computing, and a reference system. The developer kit uses an “Ampere Altra” CPU with 80 Arm “Arm Neoverse” processor cores running at up to 3.3GHz and two “NVIDIA A100” GPUs. Each A100 has a performance of 312 teraflops in the benchmark index “FP16” for deep learning. In addition, the SDK is equipped with two “NVIDIA BlueField-2” DPUs to enhance networking, storage, and security features. NVIDIA also announced that it will strengthen its partnership with Marvell and combine Marvell’s OCTEON DPU with NVIDIA’s GPUs to accelerate applications in areas such as cloud, enterprise, telecommunications carriers and edges. In the PC space, NVIDIA also announced that it will partner with MediaTek, the world’s leading supplier of Arm-based SoCs, to build a reference platform that supports Chromium, Linux, and NVIDIA’s SDK. It will bring ray tracing graphics and AI applications to new types of laptops.