Power flops
the ability of double precision floating-point computing is commonly used to measure the scientific computing ability of CPU, that is, the ability to process 64 bit floating-point data
the processor supporting avx2 can perform 16 floating-point operations in one core and one clock cycle, Also known as 16flops
CPU power = number of cores x frequency of cores x 16flops
processors supporting avx512 can perform 32 floating-point operations in one core and one clock cycle, also known as 32flops
CPU power = number of cores x frequency of cores x 32flops
2018 and 2019 are the top years of the United States
on November 12, 2018, a new issue of the global top 500 supercomputers list was released in Dallas, the United States. The "apex" of supercomputers in the United States won the championship again. The total number of supercomputers on the list in China still ranks first, which is further increased compared with the previous issue, accounting for more than 45% of the total supercomputers on the list. China's supercomputing "Shenwei · light of Taihu Lake" and "Tianhe 2" ranked third and fourth respectively
on November 19, 2019, a new issue of the global top 500 supercomputers list came out, and the US supercomputer "apex" won the championship again, while China continued to expand its leading advantage in quantity and further narrowed its gap with the US in terms of total computing power
there is no change in the top ten of the new list. The U.S. supercomputer "zenith" has reached the top again with 1.486 billion floating-point operations per second. The second place is the U.S. supercomputer "ridge". China's supercomputer "Shenwei · light of Taihu Lake" and "tianhe-2" rank third and fourth respectively
in terms of the number on the list, there are 228 super computers in China, ranking the first in the list, 9 more than half a year ago; The United States ranked second with 117. From the perspective of total calculation power, the proportion of American Super calculation is 37.1%, and that of China is 32.3%
extended data:
vertex performance data
summit supercomputer uses IBM Power9 microprocessor and NVIDIA Volta GPU for mathematical collaborative processing. Summit's predecessor, Titan supercomputer, has more than 18000 nodes, while summit will have about 3400 nodes. Each node will have at least 500GB of coherent memory and 800gb of nonvolatile memory
The original performance ofsummit supercomputer is 150 petaflops, and the delivery performance is 200 petaflops. The performance index of taihulight supercomputer in China is 93 petaflops, and the peak performance is 124.5 petaflops. IBM's supercomputer deal is said to be worth 325 million dollars
after Linx software is opened, select the calculation scale, memory usage and running times
recommended settings:
al core computing scale: 4000 runs: 1 ~ 2
four core computing scale: 8000 runs: 3
eight core computing scale: 10000 runs: 3
floating point unit description: a gflops (gigaflops) is equal to 1 billion (= 10 ^ 9) floating point operations per second
the so-called "floating-point operation" here actually includes all operations involving decimals. This kind of operation often appears in some kind of application software, and they also take more time than integer operation. Most of today's processors have a special "floating point unit" (FPU) for processing floating-point operations. Therefore, what flops measures is actually the execution speed of FPU. Linpack is one of the most commonly used benchmarks to measure flops
3. Gigaflops is one billion floating-point operations per second, which is also the unit to describe the floating-point computing capacity of computers. The mainstream CPU is generally between 20-60 gflops.
I have contacted 570595780 and other IBM minicomputers, and their performance has been greatly improved< At present, I mainly use 780 small computers, which are power7 + processors.
if we say flops
the previous generation of power7, an 8-core power7 4.1G is 264.96 gflops. Second kill Intel Core i7-980 Xe (107g)
while 7 + is more than 25% higher than 7. My configuration is 3.72g with a lower main frequency, and the performance is estimated to be similar to 4.1G of the next generation 7.
a fully equipped 780 minicomputer has 16 8-core CPUs, with a total of 4225gflops
computing power: in short, it is a quantitative indicator of the computing speed of your mining machine. For example, 1t computing power is the 12th power of 10 in 1s. If the 12th power of this 10 can work out the results that meet the conditions, it will be g. If not, it can be said that it is in vain
in the face of the exponential increase of data, computing power is the biggest demand for enterprises and institutions, and improving computing power requires higher performance CPU and GPU
last time, AMD processor pushed the computing power of HPC to billion times, but now AMD with epyc processor pushes the computing power of supercomputing to 10 billion times again. AMD's two E-class supercomputer systems, frontier and El Capitan, are scheled to be delivered in 2021 and 2023, respectively. They will achieve the expected processing performance of more than 150 exaflops (10 billion times) and 2 exaflops, respectively, and are expected to become the fastest supercomputer in the world
it is a big challenge for any manufacturer to improve the computing power in a short time. How did amd make such great progress? Let's start in 2017
in 2017, amd adopted a new Zen architecture, launched the first generation epyc processor, and dramatically increased the number of single processor cores to 32 cores. Two years later, the introction of the second generation epyc processor not only upgraded the architecture to zen2, but also reced the process technology from 14nm to 7Nm, which improved the IPC performance by 15%
compared with the Zen architecture, the new zen2 architecture optimizes the L1 instruction cache, doubles the operation cache capacity and floating point unit data bit width, and doubles the L3 cache to 16MB. The 64 core epyc processor easily has 128MB L3 cache. What's more, the second generation epyc adopts 7Nm process, which effectively reces the power consumption. Under 225W TDP, the number of cores can be increased to 64 cores, which makes its performance significantly improved
in the past year, the second generation of AMD epyc processor has achieved more than 140 world records, covering cloud computing, virtualization, high-performance computing, big data analysis and other fields, and also with strong performance to meet the increasing demand of enterprises or institutions for computing power
therefore, relying on the leading performance of epyc processor and ultra-high power consumption ratio, amd has not only won more market share and broken many world records, but also expanded its ecosystem.
Linx software can be used to test the floating-point computing ability
after Linx software is opened, select the calculation scale, memory usage and running times
recommended settings:
al core computing scale: 4000 runs: 1 ~ 2
four core computing scale: 8000 runs: 3
eight core computing scale: 10000 runs: 3
floating point unit description: a gflops (gigaflops) is equal to 1 billion (= 10 ^ 9) floating point operations per second