Nvidia Unveils First Grace CPU Superchip and Hopper H100 Data Center GPU
Nvidia introduces its first-ever discrete Grace CPU Superchip alongside the powerful Hopper H100 GPU, both designed specifically for data centers and artificial intelligence workloads rather than consumer gaming.
Nvidia CEO Jensen Huang kicks off the Graphics Technology Conference (GTC) by unveiling the company's first-ever discrete CPU, known as the Grace CPU Superchip. This new processor combines two Grace CPUs using Nvidia's NVLink technology, taking a page from Apple's M1 Ultra design. However, unlike consumer chips, the 144-core Grace Superchip focuses entirely on artificial intelligence, data science, and high-memory applications.
In addition to the Grace Superchip, Nvidia reveals its next-generation Hopper architecture through the new H100 data center GPU. Built on TSMC's N4 manufacturing process, the H100 packs 80 billion transistors and stands as the first GPU to support PCIe 5.0 and HBM3 memory. Nvidia claims that just 20 of these GPUs handle the equivalent of the entire world's internet traffic, showcasing the massive leap in data processing capabilities.
Customers access this incredible power through Nvidia's fourth-generation DGX servers, which combine eight H100 GPUs to deliver 32 petaFLOPs of AI performance. For organizations needing even more computational strength, the DGX H100 SuperPod links 32 of these servers together to provide 1 exoFLOP of AI performance and 20TB of HBM3 memory. Nvidia debuts this new architecture in its own EOS supercomputer, which harnesses 18 of these SuperPods to drive future AI research.