[ad_1]
The announcement this week of a new list of the world's 500 fastest supercomputers, unveiled at the International Conference on Supercomputers (ISC 2018) in Frankfurt, Germany, has good news and bad news for United States.
The good news is that for the first time since November 2012, the fastest computer in the world is an American machine. The Oregon Ridge National Laboratory's (ORNL) Computer Summit is now topping the list, with a sustained theoretical performance of 122.3 petaflops on the High Performance Linpack Test used to rank the list. Top500.
This machine, built by IBM, has 4,356 nodes, each equipped with two 22-core IBM Power 9 CPUs and six Nvidia Tesla V100 GPUs, all connected by a Mellanox EDR InfiniBand network.
He beat the previous winner, the Sunway TaihuLight, at the Chinese National Supercomputer Center in Wuxi, which has 93 petaflops on HPL, a score that has dominated the list since June 2016.
System No. 3 is also new, the DOE Lawrence Livermore National Laboratory (LLNR) Sierra Machine, which delivered 71.6 petaflops in HPL. Built by IBM, Sierra's architecture is quite similar to that of Summit, each of its 4,320 nodes being powered by two Power9 CPUs plus four Nvidia Tesla V100 GPUs and using the same Mellanox EDR InfiniBand as the interconnect of the system.
Summit and Sierra are part of the DOE program known as the Oak Ridge, Argonne and Lawrence Livermore (CORAL) National Collaboration Laboratories, which aims to accelerate the development of high-performance computing and accelerate exascale computing.
System No. 4 is the Tianhe-2 (or Milky Way-2) machine at the National Supercomputer Center in Guangzhou, China. This machine replaced its 5-year-old Xeon Phi accelerators with custom-built Matrix-2000 coprocessors, increasing its performance from 33.9 petaflops to 61.4 petaflops. The previous score made it the top performer on the Top500 list from June 2013 to November 2015, and was in second place before the last list was released.
The fifth-place machine is the Japanese IAI (IA Bridging Cloud Infrastructure) of the National Institute of Advanced Industrial Science and Technology (AIST), with a HPL score of 19.9 petaflops. This machine is powered by Intel Xeon Gold 20-core processors and Tesla V100 GPUs.
It is worth noting that in what is undoubtedly a better test – the High Performance Conjugate Gradient (HPCG) benchmark, which is composed of several other tests – Summit and Sierra take the first two places, followed by the computer K in Japan RIKEN Advanced Institute for Computational Science (AICS), an old system based on SPARC64VIIfx from Fujitsu.
Whatever the case may be, the good news for the United States is that they now have the right to boast of the fastest machine in the world.
The bad news is that the rest of the Top500 list is increasingly dominated by China, which now has 206 of the fastest 500 machines in the world. The United States, by contrast, is at 124 of these machines, in what may be its lowest share ever. In terms of overall performance, the United States is now leading, with 37.8% of the total performance, against 29.3% for China.
The first four systems are much faster than any other item in the list, which leads to these numbers. Note that China took the lead in the number of machines on the list six months ago, replacing the United States, which has had the largest number of systems on the list in the 25 years these systems have been in operation. followed.
It is likely that Chinese companies will now top the list of supercomputer vendors. Lenovo dominates with a 23.4% market share, followed by HPE, Inspur, Sugon and Cray. Lenovo is a Chinese company headquartered in Beijing and North Carolina. Inspur is a fast-growing Chinese server manufacturer, and Sugon is a Chinese company focused on supercomputers.
On the Green 500 list, which focuses on energy efficiency, the first three places are held by Japanese systems, starting with Shobusystem B; it is unchanged from the previous list. All three systems use Xeon main processors and PEZY-SC2 accelerators. DGX Saturn V Volta from Nvidia is in fourth place, followed by Summit.
Global lists give something to many providers. IBM, for example, built the two new US systems.
Nvidia points out that its GPU accelerators – which the company now calls "GPU Core Tensor" – are part of five of the top seven systems (all except Chinese systems); and 17 of the top 20 on the Green 500 list (all except those using Pezy accelerators). He said that the Summit had 27,648 GPUs, the largest number of computers in the world, and that they provided 95% of its energy. Nvidia also said that Summit is capable of 3 exaops (3 billion billion operations of less accuracy per second).
Meanwhile, Intel said that its processors power 95% of all Top500 systems, and 97% of all new systems, including 37 machines that use its new Xeon Scalable processors. Intel also said that the next version of its Omni-Path architecture, capable of connecting machines up to 200 Gb / s, will be released next year.
Source link