The new P3 series is based on NVIDIA's latest generation of Volta GPU - Tesla V100, the subject of Microsoft and Google's cloud products. The chip is released in May, the size and Apple Apple Watch watch die on the package of 21.1 billion Transistors. These transistors form more than 5700 processing cores, of which 640 are so-called Tensor Cores, which are specifically optimized for running AI models.
Amazon's new P3 example series has three specifications to provide one, four, and eight V100s per virtual machine, respectively. The highest two configurations are connected using NVIDIA's NVLink technology, which can be exchanged for data , The speed is much faster than the traditional way. Behind these three examples are up to 64 based on the Intel Xeon E5- 2686v4 CPU modified version of the vCPU support.
All of these chips allow the largest P3 instance to deliver up to 1 petaflop performance under certain circumstances, and in more practical cases, the series is about 14 times faster than the AWS previous generation P2.
AWS executives Matt Garman said in a statement, P2 example is already 'the current cloud to maximize the completion of machine learning' example, that is, P3 series release will only further consolidate the leading position of AWS, especially AWS two Big rivals Google and Microsoft have not yet supported NVIDIA V100 chips on their respective cloud platforms.
"AWS and NVIDIA are at the forefront of setting up the price benchmarks that other vendors in the industry can follow," said Karl Freund, director of high performance computing and deep learning at Moor Insights & Strategy.
There are two ways that enterprise customers can start using the new P3 series, they can be set manually, or use one of the two preconfigured Amazon Machine Images for the P3 series, including.
NVIDIA said that AWS customers are the first customers to use NVIDIA AI Cloud Container Registry, which includes commonly used deep learning frameworks such as TensorFlow, Caffee, CNTK and Torch.
"This will provide developers with the many features they need," said Jim McHugh, vice president and general manager of Enterprise Systems at Nine. "The idea is to attract users as much as possible and let the depth of learning be universal."
Although AWS is the first vendor to adopt NVIDIA's latest technology, other cloud providers will do the same, but he does not provide a specific timetable.
The first adopter of the P3 series was the Schrödinger LLC, a chemical simulation provider based in New York, where the company claimed that the new instance allowed them to run four times more simulations in the day.