Informing the IBM Community

Power/Nvidia combo bags $325m supercomputer order


Sequoia Computer

IBM’s OpenPower alliance with Nvidia has borne its first fruit in the form of a $325 million contract from the U.S. government’s Department of Energy to build two of the world’s fastest supercomputers.

A huge Power9 processor-based machine codenamed Summit will be installed at the Oak Ridge National Laboratory in Tennessee by 2017. It will peak at 150 to 300 petaFLOPS – an incredible 300 quadrillion calculations per second.

That’s about five times faster than the 54 petaFLOPS Intel Xeon-based Tianhe-2 at the Sun Yat-sen University in Guangzhou, China – currently the worlds fastest supercomputer. Tianhe-2 was built by domestic giant Inspur with China’s National University of Defense Technology.

A second behemoth codenamed Sierra will top out at a more modest 100 petaFLOPS and be based at the Lawrence Livermore National Laboratory in California. Both machines will also feature networking tech from Mellanox, another OpenPower member.

As it stands today, the world’s second and third fastest machines are already in situ at Oak Ridge and Lawrence Livermore. Oak Ridge’s 27 petaFLOPS Operteron-based Titan was built by Cray and also incorporates Nvidia tech. Lawrence Livermore’s Sequoia (pictured) is a Power-based  20 teraFLOPS BlueGene/Q from IBM.

Having tried out an array of processor options, it is interesting that the DoE has chosen to go down the Power/Nvidia route. When IBM announced the new model S824L server last month, its first Power System to feature GNU accelerators from its OpenPower Foundation partner, some commentators wondered what the benefit would be.

The answer, at least in the world of high-performance computing, would appear to be Nvidia’s NVLink GPU interconnect technology. This potentially allows GPUs and CPUs to share data five to 12 times faster than today and is designed, ultimately, to enable supercomputers that are 50 to 100 times faster.

The new supercomputers will also feature Nvidia’s nextgen GPU architecture, Volta, which it says will deliver considerably higher performance than its current Maxwell and subsequent Pascal designs.

Summit, like Titan, will be dedicated to open science, meaning that researchers worldwide will have the opportunity to apply for time on the system. It will also raise the bar for energy-efficient computing, providing five to ten times higher performance than Titan, while using only 10% more power.

Sierra, we are told, will be used for the National Nuclear Security Administration for, amongst other things, that most purely academic study of all – modelling the effects of intercontinental ballistic warfare.

How useful was this post?

Click on a star to rate it!

Average rating 0 / 5. Vote count: 0

No votes so far! Be the first to rate this post.