IBM, Nvidia, Dept. of Energy Team Up to Reinvent Supercomputers

As part of a multi-year effort that promises to yield advances in high-performance computing, IBM, Nvidia, and two US Department of Energy national laboratories announced this week they have created two centers of excellence for supercomputing.

Announced at the International Supercomputing Conference in Frankfurt earlier this week, the alliance calls for IBM and Nvidia to work with Lawrence Livermore National Laboratory and the Oak Ridge National Laboratory to develop a new generation of HPC technologies based on IBM OpenPower processors and graphic processor accelerators and interconnects from Nvidia.

Dave Turek, VP of technical computing at IBM, said the two centers will work on complementary projects intended to find ways to embed compute engines directly within a massively distributed network of storage systems. Rather than moving data to some central location to process it, Turek said, IBM and its partners are working on building a constellation of hardware and software technologies that would make it feasible to move compute engines into both the network and storage layer.

“As the amount of data that needs to be processed continues to grow, we need to find new approaches to memory and storage I/O,” he said. “We’re also going to need software to maintain coherence across the environment.”

The end goal, said Turek, is to make HPC systems much more broadly accessible to developers by masking as much of the complexity associated with the underlying parallelism involved in running HPC systems at an unparalleled level of scale.

Collectively, Turek said, IBM, Nvidia, and the DoE expect the downstream dividends that will be derived by those advances to have broad implications for the IT industry at large. Instead of relying on centralized computing models that revolve around a data center, the industry is on the cusp of a paradigm shift in terms of how massive amounts of data is both processed and managed.

As part of that effort, IBM and Nvidia have committed to deliver Summit and Sierra supercomputer systems to Oak Ridge and Lawrence Livermore, respectively, in 2017 with the goal of making them operational in 2018. The first prototype of the advanced supercomputers is expected to be available to developers in late 2015.

The work of the centers is managed by a technical steering group, which includes participants from IBM, Nvidia, Lawrence Livermore, Oak Ridge, and Argonne National labs.