Nvidia CEO Jen-Hsun Huang announced a new Arm-based CPU and a new generation of graphics architecture over Ampere that Nvidia says will enable the world's fastest AI supercomputer by 2023.

By leveraging Arm, a British chip design company that Nvidia is in the process of acquiring, the new Nvidia Grace CPU will dramatically increase system and memory bandwidth, Huang said.According to Nvidia, bandwidth between Grace and the Nvidia GPU will up to 900 GB/s, the company said. [It will take more than that to unseat the x86 standard in the desktop PC market, but it will be used in Nvidia's Drive Atlan autonomous vehicle SoC and the Alps supercomputer at the Swiss National Supercomputing Center by 2023.

The Alps supercomputer will be the fastest in the world for AI when completed in 2023. Alps will also use an unspecified number of "next generation Nvidia GPUs" to deliver 20 exaflops of AI computing performance. Not bad, not bad.

Even by AI inference standards, that's pretty spectacular speed. Another of NVIDIA's ventures in AI supercomputing is the Leonardo system under construction in Italy. It will use a 14,000 amp GPU and will be the world's fastest AI supercomputer after Alps. It will cost approximately 240 million euros and is expected to be completed later this year and fully operational by 2022. It is capable of 10 exaflops of AI half-precision power, with a little help from Nvidia's AI enhancements. The fastest AI supercomputer today is said to be Nvidia's 2.8 exaflops Selene.

By the way, this is a different metric than raw supercomputer speed. The world's fastest supercomputer is the Fugaku at RIKEN in Japan, which achieves over 400 petaflops on the commonly used LINPACK benchmark.

"We are thrilled to announce that the Swiss National Supercomputing Center is building a supercomputer powered by Grace and our next-generation GPUs," Huang said in his GTC keynote address. The live announcement and explanation by the cyber Jen-Hsun Huang can be seen in the replay of his GTC 2021 keynote. We're not out of Huang's kitchen yet, but as you can see, the Nvidia CEO has a few more tricks up his sleeve in his latest stream (I don't think it's his actual kitchen anymore).

The keynote is more about Nvidia's general plans for its GPU architecture than anything specifically related to gaming, but there are a few more things.

Nvidia plans to continue development of Ampere, "Ampere Next," "Ampere Next Next," and possibly a GPU architecture that drives the current RTX 30 series GPUs through 2024. by 2025, Nvidia is planning to have "Grace Next" and hopes to have "Next Next" ready. [Hopper has long been rumored to be the name of Nvidia's future GPU generation, so it will probably be incorporated into future architectures; now that Nvidia has announced the Grace CPU, a full homage to computer scientist Grace Hopper certainly has a chance to make it a reality.