NVIDIA announced their latest AI chip NVIDIA H200 this morning with faster HBM3e memory. Overall, this GPU based on Grace Hopper architecture is equipped with a memory of 141 GB at a speed of 4.8 TB/s which is faster than the previous generation NVIDIA A100.
With this the process of training LLM which is important for intelligent intelligence (AI) will also be faster. Also announced is the new NVIDIA HGX H200 platform which is an improvement to the HGX H100 system. Up to eight H200 GPUs can be attached to a GFX H200 system at one time.
The software and hardware used on the H100 are still compatible with the H200. The NVIDIA H200 GPU will be offered to customers starting Q2 2024.
Tags
GADGET