Nvidia entered a partnership with Microsoft to create one of the world’s most powerful AI supercomputers, designed to handle the massive computing workloads required to train and scale AI. Nvidia will leverage Microsoft’s scalable virtual machine instances to accelerate the advancement of generative AI models.
The AI supercomputer will be built on Microsoft’s Azure cloud infrastructure. It will include tens of thousands of Nvidia’s powerful H100 and A100 data centre GPUs and the Quantum-2 InfiniBand networking architecture. Microsoft’s Azure cloud platform combined with Nvidia’s GPUs, networking and full AI suite will enable more companies to train, deploy and scale AI, including large-scale and cutting-edge models, according to Nvidia.
The frosting on the cake will be Nvidia’s seventh-generation InfiniBand connectivity architecture. Quantum-2 devices are expected to accommodate heavy workloads with Remote Direct Memory Access (RDMA) and sophisticated acceleration engines. A bandwidth of up to 400 Gb/s will be provided in this instance. This is everything compatible with the Nvidia AI Enterprise operating system.
Nvidia said that the supercomputer could accelerate generative AI research and further advances. This is a relatively new class of large-scale language models, such as DALL-E and Stable Diffusion, that use self-learning algorithms to generate diverse content such as text, code, digital images, video and audio. In recent years, these AI models have grown rapidly, resulting in a major increase in the requirement for reliable computer infrastructure that can evolve with them.
Microsoft will optimize its DeepSpeed library to reduce computing power and memory usage during AI training workloads. It is unknown how much money these companies will invest in the supercomputer and from what date it will start working. We have to wait to find out how this project will progress and if it will finally meet its objectives.