This Monday (29), the NVIDIA announced the new DGX GH200 supercomputer targeted at Artificial Intelligence (AI) generative language applications, as well as recommender systems and data analysis workloads.
According to the company, the NVIDIA DGX supercomputer comes with NVIDIA GH200 Grace Hopper Superchips and the NVLink Switch System to combine the 256 GH200 superchips, allowing them to function as a single GPU.
The suite provides 1 exaflop of performance and 144 terabytes of shared memory, nearly 500x more memory than the previous generation NVIDIA DGX A100, released in 2020. The DGX GH200 architecture also provides 48x more NVLink bandwidth.
The GH200 superchips eliminate the need for a traditional PCIe CPU-to-GPU connection by combining an Arm-based NVIDIA Grace CPU with an NVIDIA H100 Tensor Core GPU in the same package, with interconnects via NVIDIA NVLink-C2C.
This increases the bandwidth between GPU and CPU by 7x compared to the latest PCIe technology, reduces interconnect power consumption by more than 5x, and provides a 600GB Hopper architecture GPU building block.
Google Cloud, Meta and Microsoft are among the first to gain access to the DGX GH200. NVIDIA also intends to provide the DGX GH200 design as a model for cloud service providers and other hyperscalers.
The company is also building its own DGX GH200-based AI supercomputer to power the work of its development team. Named NVIDIA Helios, which will feature four DGX GH200 systems.
Each system on the Helios supercomputer will be interconnected with the NVIDIA Quantum-2 InfiniBand network with the bandwidth of up to 400 Gb/s. The Helios is expected to include 1,024 Grace Hopper Superchips and is expected to be online by the end of this year.
