CEO Jensen Hiang made a string of bulletins throughout his Computex keynote, together with particulars about the corporate’s next DGX supercomputer. Given the place the business is clearly heading, it shouldn’t come as a shock that the DGX GH200 is largely about serving to firms develop fashions.
The supercomputer makes use of a brand new NVLink Switch System to allow 256 GH200 Grace Hopper superchips to behave as a single GPU (every of the chips has an Arm-based Grace CPU and an H100 Tensor Core GPU). This, in response to NVIDIA, permits the DGX GH200 to ship 1 exaflop of efficiency and to have 144 terabytes of shared reminiscence. The firm says that is practically 500 occasions as a lot reminiscence as you’d discover in a single DGX A100 system.
For comparability, the of the Top500 supercomputers lists as the one recognized exascale system, having reached a efficiency of practically 1.2 exaflops on the Linmark benchmark. That’s over twice the height efficiency of the second-placed system, Japan’s .
In impact, NVIDIA claims to have developed a supercomputer that may stand alongside probably the most highly effective recognized system on the planet (Meta is constructing one which it claims would be the quickest AI supercomputer on this planet as soon as it’s totally constructed out). NVIDIA says the structure of the DGX GH200 presents 10 occasions extra bandwidth than the earlier era, “delivering the facility of a large AI supercomputer with the simplicity of programming a single GPU.”
Some massive names have an interest within the DGX GH200. Google Cloud, Meta and Microsoft ought to be among the many first firms to realize entry to the supercomputer to check the way it can deal with generative AI workloads. NVIDIA says DGX GH200 supercomputers ought to be out there by the tip of 2023.
The firm is additionally constructing its personal supercomputer, Helios, that mixes 4 DGX GH200 programs. NVIDIA expects Helios to be on-line by the tip of the yr.
Huang mentioned different generative AI developments throughout his keynote, together with one on the gaming entrance. NVIDIA Avatar Cloud Engine (ACE) for Games is a service builders will have the ability to faucet into in an effort to create customized AI fashions for speech, dialog and animation. NVIDIA says ACE for Games can “give non-playable characters conversational abilities to allow them to reply to questions with lifelike personalities that evolve.”
