NVIDIA is the current leader in providing CPUs that are used in the massive generative AI servers from Microsoft and others. Today, NVIDIA announced the next generation version of its NVIDIA GH200 Grace Hopper generative AI platform, using a new (High Bandwidth Memory 3 E) processor that should make it much faster than current Grace Hopper-based servers.
NVIDIA made its announcement as part of the SIGGRAPH graphics developers conference. The company says the combination of the new Grace Hopper chip with the HBM3e processor will have over three times the memory capacity and three times the bandwidth compared to the current generation Grace Hopper platform.
NVIDIA added:
The new platform uses the Grace Hopper Superchip, which can be connected with additional Superchips by NVIDIA NVLink, allowing them to work together to deploy the giant models used for generative AI. This high-speed, coherent technology gives the GPU full access to the CPU memory, providing a combined 1.2TB of fast memory when in dual configuration.
The new HBM3e memory is supposed to be 50 percent faster than the current HBM3 memory. NVIDIA says it should offer up to 10TB/sec of combined bandwidth.
Jensen Huang, the CEO of NVIDIA, is quoted as saying:
The new GH200 Grace Hopper Superchip platform delivers this with exceptional memory technology and bandwidth to improve throughput, the ability to connect GPUs to aggregate performance without compromise, and a server design that can be easily deployed across the entire data center.
Back in May as part of Computex, NVIDIA said the current generation GH200 Grace Hopper chips would launch into full production later in 2023. The newly announced next-gen Grace Hopper chips and servers are still a ways off, with NVIDIA predicting that system builders will start delivering servers with this new platform in the second quarter of 2024.
There"s no word on pricing, but NVIDIA did state the new Grace Hopper GPUs systems "will be available in a wide range of configurations".