Microsoft has introduced the ND H100 v5 Virtual Machine series, giving enterprises an enhanced infrastructure for training and running generative AI models in Azure. Companies that sign up for the waiting list can expect the VMs starting next year.
The ND H100 v5 Virtual Machine instances, now introduced, should make it easier to train and run larger (generative) AI models, Microsoft believes. The new instances are based on Nvidia’ s H100 GPUs with its newest Hopper architecture. In total, the instances have eight of these specific GPUs on board.
In addition, the respective instances also feature the latest fourth-generation Intel Xeon Scalable processors and low-latency networking via Nvidia’s Quatum-1 CX7 InfiniBand technology.
Other features of the instances include PCIe Gen5 for a bandwidth of 64 GB per GPU and DDR5 memory for faster data transfer rates. The latter should speed up the training of the largest AI models, it is thought.
Interested companies can put themselves on a waiting list for the ND H100 v5 Virtual Machine series starting today. By next year, hundreds of thousands of instances should be available to customers who want to use them.