NVIDIA’s upcoming DGX supercomputer is all about creative AI – Nvidia

Given the direction the industry is clearly headed, it should come as no surprise that the DGX GH200 is all about helping companies develop creative AI models.

The supercomputer uses the new NVLink Switch system to allow the 256 GH200 Grace Hopper super chipset to act as a single GPU (each chip contains a Grace CPU and an H100 Tensor Core GPU). This, according to NVIDIA, allows the DGX GH200 to deliver 1 exaflop performance and has 144 TB of shared memory. The company says that’s about 500 times more memory than what you’ll find in a single DGX A100 system.

For comparison’s sake, the latest Top500 supercomputer ranking lists Frontier at Oak Ridge National Laboratory in Tennessee as the only known exascale system that achieved a performance of nearly 1.2 exaflops on the Linmark scale. This is more than twice the maximum performance of the second-place system, the Japanese Fugaku.

In fact, NVIDIA claims to have developed a supercomputer that can stand shoulder to shoulder with the most powerful known system on the planet (Meta is building a supercomputer that it claims will be the world’s fastest AI supercomputer when fully built). NVIDIA states that the DGX GH200 architecture delivers 10 times the bandwidth than the previous generation, “delivering the power of an AI supercomputer with the simplicity of GPU programming.”

Some big names are already interested in the DGX GH200. Google Cloud, Meta, and Microsoft should be among the first companies to gain access to the supercomputer to test how to handle creative AI workloads. NVIDIA says the DGX GH200 supercomputer will be available by the end of 2023.

The company also makes its own Helios supercomputer, which combines four DGX GH200 systems. NVIDIA expects Helios to be up and running by the end of the year.

Huang also touched on other developments in generative AI during his talk, including developments on the gaming front. The NVIDIA Avatar Cloud Engine (ACE) for games is a service that developers will be able to turn to in order to create custom AI models for speech, conversation, and animation. NVIDIA says ACE for games can “give NPCs conversational skills so they can answer questions as believable, well-developed characters.”