Microsoft Cognitive

NVIDIA in November, 2016 announced cooperation with Microsoft for the purpose of development of the artificial intelligence (AI) in the corporate sphere.

Thanks to the first specialized framework for problems of AI working based on GPU NVIDIA Tesla in Microsoft Azure or locally the platform is available to the companies to problems of artificial intelligence both within own IT infrastructure, and in a cloud of Microsoft now.

The GPU accelerated Cognitive tool kit from Microsoft is available in a cloud of Microsoft Azure and locally on NVIDIA DGX-1 now

Jointly the optimized platform allows to start the Microsoft Cognitive tool kit (earlier known as CNTK) on the graphic processors NVIDIA, including the NVIDIA DGX-1 supercomputer based on architecture of Pascal with NVLink technology and Azure N-Series virtual machines. Such combination provides unprecedented performance and differs in ease of use when using data for deep learning.

In only two years the number of the companies with which NVIDIA works on technologies of deep learning increased by 194 times - up to 19,000. Such industries as health care, biology, power, finance, automotive industry and production, already benefited from "deeper" a view of huge data arrays.

The Microsoft Cognitive tool kit implements and analyzes algorithms of deep learning quicker, than other available tools, effectively scaling loading in a number of environments – from central processors on graphic or even on a set of machines, at the same time without loss of accuracy of data. As a result of close cooperation of NVIDIA and Microsoft accelerated Cognitive on GPU systems and in Microsoft Azure cloud, offering the following advantages to startups and the companies:

• Broader functionality: the Cognitive tool kit allows clients to train models using one framework independently on the NVIDIA DGX-1 supercomputer or systems based on GPU NVIDIA and then to start these models in a cloud on Azure. Such scalable, hybrid approach allows the companies to prototipirovat and implement new intellectual opportunities quickly.
• Higher performance: In case of work for GPU, unlike CPU, Cognitive executes training of deep networks and inferens much quicker on the graphic processors NVIDIA available on the Azure N-Series servers and within local IT infrastructure. (1) So, for the Cognitive tools the NVIDIA DGX-1 supercomputer with architecture of Pascal and NVLink technology 170 times faster than CPU servers.

• Wider availability: Azure N-Series virtual machines based on GPU NVIDIA are available to testing only for users of Azure. In the nearest future they will become available to all users. GPU in the machines Azure can be used both for training acceleration, and for acceleration of assessment of models. Considering a huge number of the clients testing a system at present GPU Tesla in Azure N-Series virtual machines receive loading from the companies of the most different level and scale of business.