The cloud wars are heating up for high-dollar GPU compute instances. When the Amazon AWS EC2 P3 instances came out a few weeks ago we knew we were in for a slew of new cloud instance introductions. Microsoft did not disappoint as we now have the NVIDIA Tesla V100 “Volta” generation of GPUs in the Microsoft Azure cloud. Microsoft made the announcement to coincide with Supercomputing 2017 a move trumpeted by NVIDIA at the show. e5-1620
Microsoft Azure NCv3 Instances with NVIDIA Tesla V100 Volta GPUs
Microsoft has a leading cloud compute GPU architecture. The company offers virtual machines with NVIDIA Tesla K80, M60, P40, and P100 GPUs. Adding to that stable is the Microsoft Azure NCv3 instance size with NVIDIA Tesla V100 GPUs.
Realizing the need for deep learning and AI workloads, along with HPC clustered workloads to have a high-speed interconnect, Microsoft also offers a high-speed, low-latency Infiniband (Mellanox) based interconnect for its GPU compute machines. This bucks the trend with the 19 clusters that made the November 2017 Top500 supercomputers list using 25GbE. Our take is that customers paying top dollar for Microsoft Azure NCv3 and NCv2 / ND series instance types want the high performance interconnect that Microsoft Azure offers.
One of the key benefits of using the Microsoft Azure cloud NCv2 and NCv3 instances for deep learning and AI is that one can train on data already found in Azure. One can also quickly burst to spin up and then collapse capacity as required. That is especially important if a data scientist needs a fast result.
We have heard from several companies that the NVIDIA Tesla P100 based instance type machine pools have been relatively limited. We hope that as the company adds new instances the availability gets better.
You can read more on the announcement at the official Microsoft page.