Alliance for OpenUSD

NVIDIA Grace Hopper Superchips Now Available GH200 Grace Hopper Superchip combines our first Arm-based Grace CPU with an H100 GPU Using NVIDIA NVLink-C2C interconnect at 900GB/s (7X faster than PCIe Gen5), it delivers a CPU+GPU coherent access to over 600GB of memory Ideal for giant Al and HPC applications like Gen Al, deep recommenders, and vector databases GH200 delivers order-of-magnitude speed-ups compared to x86 CPUs 9X for vector databases 12X for deep recommender inference 284X for LLM (65B parameter) inference Available now from leading server manufacturers NVIDIA and SoftBank are collaborating on a platform based on GH200 for Gen Al and 5G/6G NVIDIA® 1111: 11 BOA

View entire presentation