GPUs Everywhere: New Instances in Microsoft Azure Cloud

by Ian Buck

Simulating protein molecules. Self-driving cars. Seismic exploration. Researchers and businesses have many choices to tackle these challenges with NVIDIA GPU computing.

The NVIDIA accelerated computing platform is accessible for the most demanding workloads from data centers or in the cloud. Today, researchers have another great option: GPU computing in the cloud with Microsoft Azure’s new instances.

Whether they’re sequencing DNA or providing real-time language translation, people using GPU computing in the cloud can accelerate their work, and scale it up or down on demand.

The City of Hope in Los Angeles is one such example. A team of computer scientists led by Dr. Nagarajan Vaidehi, director of the Computational Therapeutics Core at the medical research and treatment center, performs molecular modeling to better understand diseases such as cancer and diabetes.

To design drugs, Vaidehi’s team screens millions of protein molecules in 3D and performs related calculations to understand the shapes of specific candidates.

While the researchers focus their science at the molecular level of life, and the drugs that might sustain it, they’re increasingly putting their data in the cloud. Using Microsoft Azure virtual machines with NVIDIA Tesla GPU accelerators, they can scale up their computing needs to handle bigger simulations faster — and trim simulation times from weeks to days.

By using GPU resources in Azure, we can run simulations in days that would take a month on CPU-based machines.

— Nagarajan Vaidehi, City of Hope.

Accelerated Computing in the Cloud

Fighting disease is just one example of how Microsoft Azure customers use GPUs in the cloud. Other customers handle everything from high-performance compute workloads for DNA sequencing to rendering visual effects for Hollywood blockbusters.

For traditional high performance computing, Microsoft is offering its new Microsoft Azure NC Instances, powered by our Pascal architecture-based NVIDIA Tesla P100 GPU accelerators. The new instances provide double the performance of the current generation. This is huge for customers like City of Hope, because the quicker they can perform their simulations, the more progress they can make toward effectively treating diseases.

And with the explosive growth of AI and deep learning, customers are now training neural networks for everything from natural language processing to autonomous vehicles. To meet this increased demand, Microsoft is adding new ND Instances with NVIDIA Tesla P40 GPU accelerators. These offer more than double the performance over the previous generations for workloads using Microsoft’s Cognitive Toolkit, TensorFlow and other deep learning frameworks.

With ND Instances, people can work with bigger models, thanks to the 24GB of memory on each Tesla P40, and they run large-scale training and inference jobs across hundreds of GPUs.

“The power of Azure virtual machines combined with NVIDIA’s GPU accelerators enables massive scale and speed across the most performance-intensive workloads,” said Corey Sanders, director of compute at Azure. “With NVIDIA, we are using cutting-edge technology to further our mission — helping our customers achieve more with our cloud.”

The new Azure instances will be available in preview later in the year. Customers can sign up here.