As announced by the companies, the supercomputer will be integrated into Microsoft’s Azure services marketing infrastructure. The architecture is primarily designed for AI computations. The system will be equipped with tens of thousands of Hopper H100 units, which Nvidia recently launched specifically for deep learning tasks.
In addition to the H100, A100-class GPUs will also be installed. These networks are interconnected via the InfiniBand Quantum-2 network platform, so many servers can be connected to form a high-performance cluster, which communicate with each other in their data center with a bandwidth of 400 Gbps. So the computer must advance to the highest class in the world.
Further expansion planned
Microsoft contributes a significant portion of the software to the project. This includes Azure cloud infrastructure and ND and NC series virtual machines. DeepSpeed, optimization software for deep learning projects that users can work with, also comes from Redmond.
Once in cloud computer Nvidia and Microsoft’s online customers, Nvidia said, can deploy thousands of GPUs in a virtual cluster to “train even the most massive language models, build the most complex recommendation systems at scale, and enable generative AI at scale.” The companies did not provide any information about when the new supercomputer will be ready. However, the project should mark the beginning of “several years of collaboration”. Thus, cloud computing capacity will likely continue to expand over time.
“Subtly charming coffee scholar. General zombie junkie. Introvert. Alcohol nerd. Travel lover. Twitter specialist. Freelance student.”