THE 49TH EDITION OF the Top 500, the definitive run-down of the most powerful supercomputers on Earth, has been published and it contains some surprises.
Most notably, the US has been knocked out of the top three by Swiss machine the Piz Daint, which recently had an update to its Cray XC50.
The machine, based at the Swiss National Computing Centre, received some extra Nvidia Tesla P100 GPUs and thereby doubled its performance compared to the last time it was analysed, with a new benchmark of 19.6 petaflops based on Benchmarks by Linpack.
At the top of the tree is the Sunway TaihuLight, which is hosted at the National Supercomputing Centre in Wuxi, China. It stays at the top of the chart with a mind-blowing 93 petaflops, almost three times its next nearest rival - the previous number one, the Tianhe 2 (Milky Way-2) at the (other) National Supercomputer Centre at Guangzho with a score of 33.9 petaflops.
Oak Ridge National Laboratory's Titan, a Cray XK7, has dropped to number four. It has been installed for five years and not moved from the 17.6 petaflop mark.
This is only the second time this century that the US doesn't own one of the top-three computers, although it does still have five out of the top 10.
Chipwise, Intel microrprocessor can be found in 464 of the top 500 machines, with Power PC in 21 and AMD in just six. Co-processors are becoming more commonplace, with Nvidia GPUs present in 74 systems.
HPE has the most entries in the Top 500, but it's Cray, who only got round to opening an office in the UK last year, that arguably has the best performance overal. NRCPC is at number three, largely as it powers the TaihuLight.
The rest of the top 10 is as follows:
5. Sequoia: 17.2 petaflops (IBM Bluegene);
6. Cori: 14 petaflops (Cray XC40);
7. Oakforest-PACS: 13.6 petaflops (Fujitsu);
8. Fujitsu: K 10.5 petaflops (Fujitsu);
9. Mira: 8.6 petaflops (IBM Bluegene);
10. Trinity: 8.1 petaflops (Cray XC40). µ
Because some people still love to carry small pieces of cardboard
Samsung flagship might also sport enhanced facial recognition tech
Firm claims it's fixed, but customers are still reporting issues
And its called it Google Podcasts, because of course