Microsoft is creating a supercomputer cluster for the Azure cloud computing platform to assist OpenAI in developing and training ChatGPT chatbots. Microsoft has also announced the launch of a new AI virtual machine, the ND H100 v5 VM, which uses NVIDIA’s H100 GPU (an upgrade from the A100 GPU). The virtual machine features eight NVIDIA H100 Tensor Core GPUs, each with 400 Gb/s of NVIDIA Quantum-2 CX7 InfiniBand via the next-generation NVSwitch and NVLink 4.0 interconnects.
Microsoft says it is continuing to add GPUs and expand the network, and is attempting to keep them running 24/7 with cooling systems, backup generators, and uninterruptible power supplies. To train large language models, the compute workload is distributed across thousands of GPU clusters, with the InfiniBand network accelerating this process. Microsoft will allow third parties to access ChatGPT technology running on this virtual machine through Azure OpenAI.