Nvidia Gets Aboard the Open Compute Project
Nvidia is joining the Open Compute Project with its HGX-1 hyperscale GPU accelerator, which is made of eight Tesla P100 GPUs. The Tesla P100 is based on Nvidia’s Pascal architecture, the same which was used for advanced graphics cards like the GeForce GTX 1080 Ti. The HGX-1 will allow machines to process faster and more efficiently and thereby make them learn faster.
The Open Compute Project is a cooperative venture by industry giants like Google, Microsoft and Facebook, which aims to create better and smarter servers for use by these companies as well as everyone else. Though it might go too far to call the project truly open source, the designs are shared with the public in the hopes of speeding along a redesign of server hardware to create maximum efficiency. Nvidia is joining the project as a partner of Microsoft’s and the HGX-1 will likely be used as part of the Olympus server design.
According to Nvidia, the company is joining the OCP because it saw an opportunity to help set the standard for cloud computing. The HGX-1 is a good candidate for that as the accelerator should prove very powerful; though it fits “only” eight GPUs due to design limitations, up to four of them can be attached too a single Olympus server, giving these new machines unprecedented power.
Nvidia seems to see a very definite role for itself on deep-learning side of the market: yesterday the company announced the Jetson TX2, a chip for use with embedded AI in small IoT devices, while the HGX-1 will be able to speed along machine learning for servers. At the same time, however, Nvidia does not stand to gain directly from the GPU accelerator, showing that the company may just be along for the thrilling ride.
Fergus has been tinkering with computers since he was a kid and likes to put a stop to parties by listing the specs of all the digital devices in the room. It's best not to let him near your computer since he'll take it apart and may not put it back together again before he leaves.