Nine of the world’s fastest GPU supercomputers
Big and beefy heterogeneous HPC systems
When it comes to groundbreaking innovations in supercomputing over the last decade, there are few shifts that made bigger waves than the introduction of heterogeneous computing.
Coupling high-performance host processors with a graphic-processing boost has helped several top-ranked supercomputers achieve performance and energy efficiency results with GPU computing.
In addition to a thriving CUDA development ecosystem, which has pushed GPU computing to an ever-growing list of systems and applications, GPU based supercomputers keep pushing forward, most recently, with the addition to the NVIDIA Tesla K80 accelerators. On this year’s Top 500 list of the world’s fastest supercomputers, 52 systems were outfitted with NVIDIA GPUs, adding to the countless other HPC machines that do not run the benchmark for list placement at major companies and research institutions alike.
To highlight this progress, below, ranked in order of theoretical peak performance, are ten of the fastest GPU enabled supercomputers on the planet.
1. Titan Supercomputer at Oak Ridge National Lab
Titan, a Cray XK7 system installed at the Department of Energy’s Oak Ridge National Laboratory remains the number two system. It is an extensive upgrade to the former top-ranked Jaguar supercomputer, a process that started in 2011 and was complete in 2012. Titan held the top spot on the Top 500 in November 2012 before the Tianhe machine was benchmarked. It achieved 17.59 Pflop/s on the Linpack benchmark using 261,632 of its NVIDIA K20x accelerator cores. Titan is one of the most energy efficient systems on the list consuming a total of 8.21 MW and delivering 2.143 Gflops/W. Titan’s successor, the Summit supercomputer, will sport an entirely different architecture (NVIDIA/IBM/Mellanox) when it appears at Oak Ridge sometime in 2018 and is likely to also appear in the top five (if we were to make a guess—few saw Tianhe-1 or its successor coming, to be fair).
2. “Piz Daint” at the Swiss National Supercomputing Centre
At No. 6 is Piz Daint, a Cray XC30 system installed at the Swiss National Supercomputing Centre (CSCS) in Lugano, Switzerland and the most powerful system in Europe. Piz Daint achieved 6.27 Pflop/s on the Linpack benchmark using 73,808 NVIDIA K20x accelerator cores. Piz Daint is also the most energy efficient system in the top ten of the Top 500 list, consuming a total of 2.33 MW and delivering 2.7 Gflops/W.
3. Tulip Trading Supercomputer in Australia
Although Tulip Trading might not be a household name, this Australian supercomputer, named C01N, places at #15 on the list of the world’s fastest supercomputers. This SGI ICE X/Superblade system is outfitted with 265,440 CPU cores and is capable of 3.5 petaflops of peak performance. Interestingly, unlike other top-ranked supercomputers that leverage GPUs for computational boosts, this system makes uses of the NVIDIA M2090 accelerators which, while not new processors, are clearly powering some of the world’s fastest trades and adding to the overall performance of one of the few commercial systems in the top 20.
4. Eni S.p.A Oil and Gas Supercomputer in Italy
This machine for oil and gas exploration at one of Europe’s largest energy resource discovery firms is ranked at #17 on the Top 500 list of the fastest supercomputers. NVIDIA K20 GPUs accelerate the iDataplex machine from IBM, offering a total peak performance of 3.1 petaflops of peak performance.
This is one of several oil and gas supercomputers on the Top 500 that use GPUs to accelerate seismic modeling and other energy resource discovery applications.