DENVER — Nvidia plans to bring supercomputing to the masses thanks to a collaboration with Microsoft Azure.

According to Nvidia CEO and leather jacket aficionado Jensen Huang, more and more researchers are turning to cloud-based supercomputing because it's easier than building a supercomputer cluster. He said the partnership will enable researchers from all over the world to rent supercomputing time using Microsoft Azure. The largest deployments available will place up to 800 Nvidia V100 Tensor core GPUs in the hands of Azure customers.

"Until now, access to supercomputers for AI [artificial intelligence] and high-performance computing has been reserved for the world's largest businesses and organizations," said Ian Buck, VP and general manager of accelerated computing at Nvidia, in a statement. "Microsoft Azure's new offering democratizes AI, giving wide access to an essential tool needed to solve some of the world's biggest challenges."

Key to this equation, however, is making GPU-accelerated workloads cost-effective.

Huang said while the price per hour for a GPU compute remains higher than CPU instances, the cost to actually complete a job using a GPU often ends up costing significantly less. He said this is because a GPU can often complete the task in a fraction of the time.

"Even when you use multiple GPUs on one of the most powerful, and therefore most expensive instances in the cloud, is actually the most affordable," he said. "The best way to reduce cost is to get your job done and get off."

Nvidia also made good on a promise to bring its Cuda-X software platform to Arm this week with the unveiling of an Arm-based, GPU-accelerated server. The reference design features four Nvidia Volta GPUs and is designed to work seamlessly with any manufacturer's CPUs.

With this new integration, Nvidia aims to address growing demand within the high performance computing (HPC) community for a more diverse set of CPU architectures.

But bringing Cuda to Arm didn't happen overnight. Nvidia collaborated with dozens of industry leaders including chip makers, software developers, and researchers to prove the technology could work.

The reference design involved partnerships with several chipmakers including Ampere, Fujitsu, and Marvell. Nvidia also worked closely with Cray and its parent company Hewlett Packard Enterprise (HPE), which were two of the first to offer Arm servers.

Nvidia also worked with Linux distributors including Canonical, Red Hat and SUSE as well as several HPC software developers to ensure applications like GROMACS, LAMMPS, MILC, NAMD, Quantum Expresso, and Relion will work on the new combination.

"The industry has been really really fantastic," Huang said. "Everyone is jumping on it, and we already have 30 applications."