Built to handle demanding AI, machine learning and high-performance computing applications, this new offering can supply up to 800 interconnected NVIDIA Tesla V100 chips that are perfectly at home with any kind of deep learning workloads via the cloud.
As the chip giant indicates, this new instance enables dramatic performance benefits and cost advantages compared to traditional CPU-based computing.
In other words, the possibility of renting an entire AI supercomputer on demand has just become real, as Ian Buck, vice president and general manager of accelerated computing at NVIDIA puts it:
“Until now, access to supercomputers for AI and high-performance computing has been reserved for the world’s largest businesses and organizations. Microsoft Azure’s new offering democratizes AI, giving wide access to an essential tool needed to solve some of the world’s biggest challenges.”
As an example, the company says that AI researchers will now have the luxury to spin up multiple NDv2 instances and train complex conversational AI models in mere hours. More importantly, this performance also scales linearly to a hundred instances, which makes it ideal for large-scale simulations.
NVIDIA and Microsoft engineers joined hands to build this scalable GPU-accelerated supercomputer, using 64 instances on a prerelease version of the cluster.
CEO Jensen Huang shared the news today onstage at SC19, a supercomputing conference in Denver.
And it comes the same day as AWS unveiled plans to launch some of its most powerful cloud EC2 instances ever powered by AMD EPYC Rome processors, and a day after Intel made official its Pone Vecchio GPU for datacenters.
NDv2 is available now in preview.