Rent a deep learning
Scan in the UK plans to offer the Nvidia DGX-1 deep learning super computer for buy or even a rent. This is so far a unique approach as many deep learning startups will get a chance to try the system before they buy it. And then there are many Universities who would love to get their hands on 170 teraFlops of deep learning performance.
Fudzilla had a chance to chat with Lee Rositer, who is in charge of the professional deep learning market and the customers. Since this market started only three years ago, this is the right time to jump on it as the research led by key universities, key companies including Microsoft, Google, Facebook and Baidu see a great potential in the deep learning market.
Nvidia managed to get even very conservative companies such as SAP excited about deep learning as jut as you can recognize and optimized images and speech, you will be able to use the same technology to improve processes in a company.
Scan.co.uk has realized that it is a good idea to jump into this market as the education market, startups, oil and gas industry as well as financial market are looking into using deep learning computers.
Scan will offer a bundle of solution for early adopters and the one that caught our attention was the ability to rent a DXG-1 System with eight P100 powered by Pascal chips and 16GB HBM 2 memory, summing up to a total horsepower of 28672 Cuda cores or 170 teraFLOPS of half-precision (FP16) peak performance.
Nvidia DGX-1 should be one of the fastest deep learning supercomputers on the market and they will sell for more than $ 100,000 even to top educational clients. The rest would probably end up paying even more.
Deep learning uses parallel process abilities of Nvidia GPUs to reproduce the functions of a brain in a deep neural network in order to detect patterns and concepts in data. This process, due its incredible ability to parallelize, can dramatically speed the speech recognition, language translating, image recognition and trend analysis.
There is a lot of information at the Scan website.
Autore: Fudzilla.com – Home
