In its latest release, NVIDIA announced a bunch of new additions to its high-performance computing (HPC) platform. The platform constitutes a full technology stack with CPUs, GPUs, DPUs, systems, networking, and a range of AI and HPC softwares. The platform is available for researchers to push forward their work on cutting-edge technology systems, on-premises or the cloud.  

The announcements also include the wide adoption of its next-generation H100 Tensor Core GPUs and Quantum-2 Infiniband, along with providing new offerings on Microsoft Azure and 50+ new partnerships to advance scientific discovery. 

The company made the announcements at SC22 where it also released major updates to its cuQuantum, CUDA® and BlueField® DOCA™ acceleration libraries, while also announcing support for its Omniverse™ simulation platform on NVIDIA A100- and H100-powered systems. 

Only a few days ago, researchers from multiple institutions, including NVIDIA, were able to develop a language model that used NVIDIA’s high performance computing systems to predict new and emergent variants of several pandemic-causing viruses. 

Jensen Huang, CEO of NVIDIA, in discussing the updates NVIDIA is bringing, spoke about the service AI is doing to scientific research, while expressing how providing a universal scientific computing platform, that accelerates both numerical and AI methods, can provide researchers with the technology to propel scientific discoveries. 

Microsoft Azure adoption of Quantum-2 Infiniband 

At the GPU Technology Conference in March 2022, NVIDIA announced that the Quantum-2, its cloud-native supercomputing platform, will be open for general availability. Microsoft’s incorporation of Quantum-2 in its next generation Azure HPC is aimed at maximising performance and cutting costs for its HPC customers.  

Quantum-2, built with the Infiniband architecture, empowers leading supercomputer data centres with HPC solutions by:

• Extending its in-network computing with preconfigured and programmable compute engines to enable analysing data and perform complex simulations with increased speed and efficiency

• Delivering performance isolation by proactive monitoring and congestion management, as well as eliminating performance jitter to bring predictability to the performance

Azure will use 400 Gigabit NVIDIA Quantum-2 InfiniBand for the offload of MPI collectives, enhanced congestion control, and enhanced adaptive routing capabilities. 

Updates to Acceleration Libraries

The updates to its cuQuantum, CUDA® and BlueField® DOCA™ acceleration libraries include:

• Addition of a multi-node, multi-GPU Eigensolver to the NVIDIA CUDA library for unparalleled scale and performance in HPC applications like VASP, a package for first-principles quantum mechanical solutions

• NVIDIA cuQuantum software development kit will support approximate tensor network methods enabling researchers to simulate tens of thousands of qubits, while also automatically enabling multi-node, multi-GPU support for quantum simulation using the cuQuantum Appliance

• Bluefield DPUs will integrate NVIDIA DOCA, the open cloud SDK and acceleration framework to provide advanced programmability, security and functionality, opening room for strong use cases. 

With the NVIDIA HPC acceleration libraries, researchers can scale across multiple servers, which will in turn drive huge performance boosts for scientific research. The libraries are available on all leading cloud platforms like AWS, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure.