The new Green500 list of the most energy-efficient supercomputers has demonstrated that the top supercomputers have made large leaps in energy efficiency over the last 12 months. This is largely due to the presence of new technologies taking advantage of the latest semiconductor technologies in the HPC space, as new manycore processors from Intel and NVIDIA begin to penetrate the top systems and overall performance per watt rises considerably.
The DGX SATURNV, NVIDIA’s first petascale, in-house supercomputer, earned the top spot on the Green500 list with a rating of 9.46 gigaflops per watt. This represents a huge increase of more than 40 per cent power/watt energy efficiency from the number one system just six months ago on the June 2016. Much of this increase comes from the use new HPC technologies such as NVIDIA’s DGX-1 which houses eight P100 GPUs. DGX SATURNV is comprised of 125 NVIDIA Pascal-powered DGX-1 server nodes. This creates a supercomputer with 1,000 NVIDIA Tesla P100 accelerators, coupled with NVIDIA’s deep learning software stack providing high-performance aimed firmly at large-scale, deep learning applications.
The newly upgraded Piz Daint system based on a CrayXC50 the newest version of Cray’s XC series supercomputing platform. The SC series includes all of Cray’s latest technologies such as Cray’s Aries network interconnect, DataWarp I/O accelerator technology and support for the latest Intel and Nvidia processors including the P100.
The Swiss centre CSCS which houses Piz Daint recently upgraded two if its systems, creating one of the most powerful supercomputers in the world. It combined the original ‘Piz Daint’ based on a Cray XC30 with the Centre’s Cray XC40 supercomputer nick-named ‘Piz Dora’.
Speaking prior to the upgrade Professor Thomas Shulthess, director of the Swiss National Supercomputing Centre (CSCS) said: ‘Our new Cray XC50 supercomputer will significantly accelerate our computational research capabilities allowing our users to perform more advanced, data-intensive simulations, visualisations and data analyses across a wide array of scientific studies. Cray’s next-generation supercomputer and its continued integration of GPU acceleration has created a powerful and efficient hybrid multi-core system for addressing our current and future HPC workloads’.
Piz Daint took second place with a mark of 7.45 gigaflops per watt. That system rose up the list this year due to the new upgrade which replaced its original K20x graphics processors to the new P100 GPUs. Although it’s based on the same processor technology as the DGX SATURNV, it uses a lower ratio of GPUs to CPUs.
Meanwhile, the former number one Green500 system, the Shoubu supercomputer, equipped with the PEZY-SCnp manycore accelerator, slipped into third place with an energy efficiency of 6.67 gigaflops per watt as Nvidia took the top spots with large installations of its latest accelerator hardware.
With the exception of the XStream system at number eight, which uses the NVIDIA K80 accelerator, the remainder of the top 10 systems rely on homogeneous manycore processors. Five of those machines are powered by Intel’s new ‘Knights Landing’ Xeon Phi processors, which have been report to provide energy efficiency of between 3.84 and 5.61 gigaflops per watt. The main outlier here is the Sunway TaihuLight supercomputer, the number one system on the TOP500 list, which earned the number four spot on the Green500 with a value of 6.05 gigaflops per watt. Taihulight is powered by the SW26010, a 260-core ShenWei processor, designed and manufactured in China.
Heterogeneous systems continue to demonstrate better energy efficiency than homogeneous systems, but that gap appears to be narrowing with the appearance of standalone, manycore processors like Xeon Phi and ShenWei. However, the question of whether these manycore architectures will be able to catch or overtake coprocessors still remains.
A lot of advances in energy efficiency on the Green500 this year have been due to the release of new generations of processor technology, such as Intel’s ‘Knights Landing’ and Nvidia’s P100. As such it is likely that there will not be as large a jump in efficiency in the next twelve months. The next big jump is apt to occur when the next generation of GPUs, Xeon Phi processors, and perhaps the Chinese manycore processors are available to supercomputer makers in the late 2017/early 2018 timeframe.
If we use the most efficient system on the current list – the DGX SATURNV – as a baseline, an exascale system with that level of energy efficiency would need approximately 105 MW. That is about five times the desired goal of 20 MW, but only a little more than three times the 30 MW upper limit of the initial systems targeted under the US exascale effort. The Top500 news recently reported that Green500 founder Wuchun Feng is optimistic that the numbers are starting to line up: ‘While reaching a 20 MW exascale is highly unlikely by 2020, China has unofficially indicated that their target power for an exascale supercomputer in 2020 to be 40 MW. This 40 MW projection, I think, is attainable.’
The next Green500 list will be announced in June 2017 in conjunction with the ISC High Performance conference, which will take place in Frankfurt, Germany.
