IBM and Nvidia to produce world's fastest supercomputers

The US Department of Energy has awarded a $325 million contract to Nvidia and IBM to build two GPU-accelerated supercomputers which are set to be delivered by 2017.

Once operational, both will comfortably take away the top two slots from the current HPC champions, Oak Ridge's "Titan" and China's Tianhe-2.

The faster model, "Summit", will replace Titan at the Oak Ridge National Laboratory and will deliver between 150 and 300 peak petaflops while its smaller counterpart, "Sierra" will be based at the Lawrence Livermore National Laboratory and hit in excess of 100 peak petaflops. That's more than the current top two supercomputers combined.

The former will be used for "open science" while the latter will be a key element of the US national nuclear security mission. IBM provided the Power CPU while Nvidia brought in its GPU as well as its NV-Link interface.

Only a few weeks ago, UK's MET Office purchased a Cray supercomputer for £97 million, which doesn't sound much of a deal given that it reaches "only" 16 peak petaflops.

But then, it did come with quite a lot of storage and system memory (although we're still waiting on the MET Office to get back to us on what looks like arithmetic mistakes).