Nvidia has recently announced its latest high-end chip, the GH200 Grace Hopper Superchip, which is designed for training AI models at giant scale.
The GH200 is a breakthrough accelerated CPU that combines the NVIDIA Grace™ and Hopper™ architectures using NVIDIA® NVLink®-C2C to deliver a CPU+GPU coherent memory model for AI and HPC applications. The superchip delivers up to 10X higher performance for applications running terabytes of data, enabling scientists and researchers to reach unprecedented solutions for the world’s most complex problems.
The technical bit
The GH200 features 72 cores of Grace CPU outfitted with 480 GB of ECC LPDDR5X memory, as well as the GH100 compute GPU that is paired with 141 GB of HBM3E memory that comes in six 24 GB stacks and uses a 6,144-bit memory interface.
The GH200 also has a new 900 gigabytes per second (GB/s) coherent interface, which is 7X faster than PCIe Gen5, and supercharges accelerated computing and generative AI with HBM3 and HBM3e GPU memory. The GH200 can run all NVIDIA software stacks and platforms, including NVIDIA AI Enterprise, HPC SDK, and Omniverse™.
The GH200 is available as part of the NVIDIA DGX GH200, a massive memory supercomputer that fully connects 256 GH200 Superchips into a singular GPU. The DGX GH200 offers 144 terabytes (TB) of shared memory with linear scalability for giant AI models.
The DGX GH200 is a turnkey data centre-class solution that includes integrated software and white-glove services from NVIDIA, from design to deployment, to speed the ROI of AI.
The DGX GH200 is the only AI supercomputer that offers a massive, shared memory space of 144TB across 256 NVIDIA Grace Hopper Superchips, providing developers with nearly 500X more memory to build giant models.