nsdp + 449 eh October 31, 2021 5 hours ago, Starschy said: China outperformed The US in Supercomputing. Since March 21 two Exascale Computer are working over there. The Chinese did not make a sound and hide that Computers. Not even registered them for No 1 at Supercomputing 500. The are very quick about 8 times faster as US No 1 Oak Ridge Lab 3 times quicker as the Fugaku actual No 1 in Supercomputing The fastest European solution is outperformed by a factor 25 Not to mention from 500 Supercomputing solution near 200 are from China and about 120 from the USA. Get you facts right. Fugaku remains the No. 1 system. It has 7,630,848 cores which allowed it to achieve an HPL benchmark score of 442 Pflop/s. This puts it 3x ahead of the No. 2 system in the list. Summit, an IBM-built system at the Oak Ridge National Laboratory (ORNL) in Tennessee, USA, remains the fastest system in the U.S. and at the No. 2 spot worldwide with a performance of 148.8 Pflop/s on the HPL benchmark, which is used to rank the TOP500 list. Sierra, a system at the Lawrence Livermore National Laboratory, CA, USA is at No. 3. Its architecture is very similar to the #2 system Summit. It is built with 4,320 nodes with two Power9 CPUs and four NVIDIA Tesla V100 GPUs. Sierra achieved 94.6 Pflop/s. Sunway TaihuLight, a system developed by China's National Research Center of Parallel Computer Engineering & Technology (NRCPC) and installed at the National Supercomputing Center in Wuxi, which is in China's Jiangsu province, is listed at the No. 4 position with 93 Pflop/s. Perlmutter at No. 5 is new in the TOP10. It is based on the HPE Cray "Shasta" platform, and a heterogeneous system with AMD EPYC based nodes and 1536 NVIDIA A100 accelerated nodes. Perlmutter achieved 64.6 Pflop/s. Selene, now at No. 6, is an NVIDIA DGX A100 SuperPOD installed inhouse at NVIDIA in the USA. The system is based on an AMD EPYC processor with NVIDIA A100 for acceleration and a Mellanox HDR InfiniBand as a network and achieved 63.4 https://www.top500.org/lists/top500/2021/06/Pflop/s Frontier at Oak Ridge will be the new number 1 at 1.5 exaflops whem scores are announced at Super Computing in two weeks. That is 3.5 times the Fuqku rank . That is with 100 cabinets. there is room and power for 100 more and El Capitan will complete early next year with Genoa instead of Milano cpus. Frontier will be announced as both number 1 in top 500 and Green 500. Quote Share this post Link to post Share on other sites
nsdp + 449 eh October 31, 2021 6 hours ago, Andrei Moutchkine said: Top500 is a competition in government pork barrel, good for running supercomputing benchmarks and nothing else. All the practical supercomputers fit into a single box. Depends on what you are doing. If you play tiddle dewinks that is true. But if you are dispatching 75 gw with 50,000 miles of transmission 60,000 OCB pluse keep power flows withing limits on each circuit you will need more than your toy box. How many stock trades will your toy box handlles and how many hours will your ATM users be waiting. , Quote Share this post Link to post Share on other sites
SUZNV + 1,197 October 31, 2021 11 hours ago, Andrei Moutchkine said: EUV lithography is not really in full production. Only for critical bits. What does Trump got to say over technology that is not even American? (there is one Dutch and one Japanese provider, AFAIK) ARM patents are beyond stale. All they own is the ISA. You are as confused as any American about there being any freedom in US. There is much advertising to the effect, but it is not really the same as having it. I don't approve of the Chinese policy of sheltering their citizens. Russian one is better - let the citizens work up the resistance to the toxic Western media. Sheltering their citizenry like that was one of the most significant, if not most significant, contribution to the Soviet demise, IMHO. EUV tech has been used in making chips for years, but the machine are very complicated in manufacture, need lots of hand on experience for deployment and operation. ARM ISA is for writing instructions for chips. ARM charges IP license on producing ARM architecture and chip customizations. I have friends from hi school whose family was sponsored to the US to work in chip industry and my younger brother worked in AMD in Singapore. Funny about your assumption about me as any American. I am neither here nor there. I grew up in a communism country, studied and live in a progressive country 15 years and now living in the US. I can tell the differences between Marxism, Leninism, Stalinism, Maoism and so called progressivism/liberalism/socialism , along side with Capitalism and individualism in the old US. The social conflicts between young and old generations in Wester Countries. I am also familiar with mainstream propaganda in both Western and Eastern countries and the techniques they use. Western mainstream use these techniques poorly yet maybe many Westerner didn't have a vaccine for propaganda when they were young and too many fell for entitlement and divisiveness. Many US people are too busy with life and depends on mainstream for not only information but the interpretation as well and therefore, kind of ignorance about what happened in the world out of US. I even somewhat agree with Putin's view about Western media. I am not that against dictatorship, as Singapore with Lee Kuang Yew and Korea with Park Chung He but the bad part about lifetime dictatorship is the transition of power to the next and people may pay for the consequence they didn't vote for. In the Democracy countries, at least half of the voters are responsible for the consequence in any election, for better or worse. I am not fan of capital gain tax and property tax should be viewed as unconstitutional in US people' eyes, as people pay for the property with after taxed money and loan. Ironically Communism countries such as China and Vietnam don't have residential property tax but Xi may make one soon to deflate the real estate bubble and inequality. To be fair, this problem has been the cancer long before Xi came to power. But at least Rich people in China move to Western Countries (which inflate asset bubble in Big cities in UK, US, Australia, NZ, Canada... ) and not the other way round so the talents would want the same. China economy didn't have any correction the Cold War end and geopolitically they have to import raw material and oil, depends on export while the domestic in the edge with real estate bubble, which makes China vulnerable for their first recession in 30 years, somewhat similar with Japan 30 years ago. Quote Share this post Link to post Share on other sites
Andrei Moutchkine + 828 October 31, 2021 6 hours ago, nsdp said: Depends on what you are doing. If you play tiddle dewinks that is true. But if you are dispatching 75 gw with 50,000 miles of transmission 60,000 OCB pluse keep power flows withing limits on each circuit you will need more than your toy box. How many stock trades will your toy box handlles and how many hours will your ATM users be waiting. , My toy box will handle more trades than any machine in the Top500, because it runs Q/kdb https://kx.com/ You need a single box (or two, in case of HP ThunderDome) because you need an illusion of a unified memory, because all the Top500 machines do is orchestrating internode communications. https://en.wikipedia.org/wiki/Amdahl's_law Quote Share this post Link to post Share on other sites
Starschy + 211 PM October 31, 2021 16 hours ago, nsdp said: Get you facts right. Fugaku remains the No. 1 system. It has 7,630,848 cores which allowed it to achieve an HPL benchmark score of 442 Pflop/s. This puts it 3x ahead of the No. 2 system in the list. Summit, an IBM-built system at the Oak Ridge National Laboratory (ORNL) in Tennessee, USA, remains the fastest system in the U.S. and at the No. 2 spot worldwide with a performance of 148.8 Pflop/s on the HPL benchmark, which is used to rank the TOP500 Frontier at Oak Ridge will be the new number 1 at 1.5 exaflops whem scores are announced at Super Computing in two weeks. That is 3.5 times the Fuqku rank . That is with 100 https://www.nextplatform.com/2021/10/26/china-has-already-reached-exascale-on-two-separate-systems/ We have qualified Information that those Chinese Systems are running. Nothing more or less. Quote Share this post Link to post Share on other sites
surrept33 + 612 st October 31, 2021 10 hours ago, Andrei Moutchkine said: My toy box will handle more trades than any machine in the Top500, because it runs Q/kdb https://kx.com/ You need a single box (or two, in case of HP ThunderDome) because you need an illusion of a unified memory, because all the Top500 machines do is orchestrating internode communications. https://en.wikipedia.org/wiki/Amdahl's_law Actual scientific discovery happens on the Top500, things that require vertical scaling (i.e, traditional HPC) vs horizontal scaling (e.g, kdb) are FAR different. Modern exascale computing requires both. From MIT's CS50 course: https://computationalthinking.mit.edu/Fall20/lecture24/#why_we_dont_game_on_supercomputers 1 Quote Share this post Link to post Share on other sites
Andrei Moutchkine + 828 October 31, 2021 (edited) 2 hours ago, surrept33 said: Actual scientific discovery happens on the Top500, things that require vertical scaling (i.e, traditional HPC) vs horizontal scaling (e.g, kdb) are FAR different. Modern exascale computing requires both. From MIT's CS50 course: https://computationalthinking.mit.edu/Fall20/lecture24/#why_we_dont_game_on_supercomputers What kind of actual discovery, give me a real example? Using your parlance, the Top500 machines are horizontal scalers. They are loosely coupled clusters on a fast interconnect, where you have to manage work distribution by explicit messaging. You need at least NUMA-level abstraction for vertical scaling. Why would I need a lower division class for non-majors? Here, some prior art to what NVidia is doing today (realtime raytracing) in yours truly class work from late 90-ties https://web.archive.org/web/20060901125459/http://www.csua.berkeley.edu/~muchandr/pert/pert.html And no, I don't claim there is any scientific validity in any of this. Edited October 31, 2021 by Andrei Moutchkine Quote Share this post Link to post Share on other sites
nsdp + 449 eh November 1, 2021 Ok smart ass. https://patents.google.com/patent/CN104937222B/en. We used 100 hours of time on Titan in Oak Ridge's Small Business Incubator program. Two key features is that it uses retired jet engines (recycle, reuse, cuts costs) and has a NEGATIVE Delta T for thermodynamics. It extracts energy from the biosphere to generate electricity and solves the issue of synchronous generation for the grid There are separate patents for mixing and burning hydrogen and oxygen in pure state without melting the turbine. This takes wind power and solar turns them into a stable synchronous source of electricity for grid and lowers ambient air temps on a global scale. You wouldn't even have the slightest clue how to solve this puzzle. Quote Share this post Link to post Share on other sites
Most Active Members
Recently Online
Popular Members