SAN JOSE, Calif.—Nvidia refreshed its lineup of GPUs for deep learning and artificial intelligence applications on Wednesday with the new 5,120-core, 7.5 teraflop Tesla V100 Volta.
The new processor is part of Nvidia's quest to come up with a new way to consistently improve computing performance in the aftermath of Moore's Law, which many industry leaders agree is pretty much dead. Instead of boosting processor speeds or cramming more transistors onto already-crowded silicon, Nvidia is championing GPU-accelerated computing, which the company's CEO Jensen Huang (pictured above) said can offer a 150 percent performance boost every year.
The Tesla V100, with a brand-new architecture called "Volta," represents that latest boost. It's the "next giant leap into the new world" of AI and high-performance computing, Huang said at Nvidia's developers conference here. The V100 will start shipping by the end of the year to data centers owned by Amazon, Microsoft, and other cloud computing providers in several different configurations.
The most common off-the-shelf version of the V100 is a $149,000 supercomputer called the DGX-1, which contains eight V100 processors. There's also a $69,000, liquid-cooled mini supercomputer called the DGX aimed at researchers who aren't using the cloud, which is powered by four V100s.
The DGX-1 made its debut last year with the V100's predecessor, the Tesla P100. Thanks to the Volta architecture, the V100-powered supercomputers offer five times more peak teraflops improvement over Pascal chips, the current-generation Nvidia GPU architecture, and 15 times more than the Maxwell architecture offers.
The V100 will power servers in the Microsoft Azure cloud and Amazon Web Services, which are the two titans of cloud offerings for deep learning. Once installed, they'll support software like Microsoft's Cognitive Toolkit and Mxnet, an open-source deep learning framework that is Amazon's preferred AI framework. Nvidia's V100 will also likely power Facebook's Caffe2 framework.
Related
- Nvidia Unveils Titan Xp, Mac Support for Pascal GPUsNvidia Unveils Titan Xp, Mac Support for Pascal GPUs
"Our strategy is to create the most productive platform for deep learning," Jensen said. He offered several demonstrations of the P100's potential applications, from an artwork algorithm that can merge the color palette of one photo (an orange sunset, for instance) with the subject of another (a beach and clouds, for instance) into an entirely new image.
Of course, the chief benefit of hosting AI processing power in the cloud is that engineers can do pretty much whatever they want with it. To that end, Nvidia also plans to launch its own GPU Cloud service, currently in an invite-only beta, which allows anyone to upload their machine algorithms to Volta servers.
"One of the best ways to enjoy deep learning is for someone else to build this incredibly complicated supercomputer on your behalf," Jensen said.