On the calculation method of float peak value

Linpack Baidu Encyclopedia: http://baike.baidu.com/view/485828.htm

Cluster theory floating-point peak =cpu frequency (GHz) x4 (CPU per clock cycle performs floating-point operations) x node number x8 (dual quad core per node)
For example, the CPU of single node machine is Xeon E5430, then its theoretical peak =1.66x4x1x8=53.12 Gflops

Another simple metric that is often used to evaluate computer performance is peak speed. The peak speed of a computer usually refers to its theoretical peak velocity, which is calculated by:

Theoretical peak speed (million times) = MHz x number of floating-point operations per clock cycle x CPUs/108

where MHz refers to the CPU's central frequency, the number of floating-point operations per clock cycle is determined by the number of floating-point units in the processor and by how many floating-point operations each floating-point unit can handle in each clock cycle. In IBM POWER4, for example, each POWER4 processor has two floating-point units, each of which can handle both an addition and a multiplication operation within a single clock cycle. So if the processor has a frequency of 1.7GHz (power4+), the processor has a peak speed of 1.7g*2*2=6.8gflop/s, or 6.8 billion times per second. For the fully configured IBM p690, the peak speed of the system is 1.7g*2*2*32=217.6gflops, or 217.6 billion times per second. Different processors can perform floating-point operations at different times per clock cycle. Common processors in the marketplace are as follows:
The number of times a floating-point operation can be performed per clock cycle:
IBM Power4 4
HPQ Pa-risc 4
HPQ Alpha 2
SUN Ultra-sparc 2
SGI MIPS 2
Intel Itanium 4
Intel Pentium 1
Intel Xeon 2

The peak speed can reflect the performance of the processor to some extent, but it does not represent the actual computing power of the computer. For this reason, many organizations try to use some standard procedure to test the computing speed of computer, in which Linpack is widely used to evaluate the actual peak computing ability of computer, which is characterized by its simple use and strong applicability. Although these indices do not reflect the whole system performance of a given system, they can be used as a correction to the peak performance of the system theory.

The world's computer TOP500 rankings are based on Linpack test results.

The theoretical flops calculation for the Intel Xeon E7-8837 is not:
2.66 (main frequency) x8 (CORE) x4 (number of floating-point operations per cycle) =85.12 Gflops
For the peak flops of AMD Opteron 6220:
3 (main frequency) x8 (CORE) x4 (number of floating-point operations per cycle) =96 Gflops
Now the puzzle is that the CPU is not able to find the number of floating-point operations per cycle. The x86 architecture currently has 4 floating-point operations/clock cycles, while Intel and AMD have certain models that can do 8 floating-point operations/clock cycles, and ask where this data can be found. Or know the post to tell the above two CPU per cycle floating-point number of times, thank you.
zouming1979 (Contact TA in station)
At present, most of the theoretical peak is calculated according to X4, because the CPU core can be a beat to execute two instructions, each instruction can perform two operations, so it is X4.
Intel's new architecture processor adds a new instruction set AVX, so one instruction can perform four operations, so the processor x8 of the new architecture means that the Intel processor that supports the new instruction set AVX is x8.
The new architecture Sandy Bridge and Ivy Bridge are x8. Server Processor e5-xxxx, e3-xxxx are x8.
The processor used on the PC is not clear and does not pay much attention.

AMD performance is too poor, and now is not too concerned about.
The memory of the wind (in-Station contact TA)
As LS says
Flops is just a theoretical value.
Although the actual AMD is high but the actual performance is not good
Involved in many aspects of CPU architecture
Recommended purchase of Intel
Psfan (Contact TA in station)
2/F: Originally Posted by zouming1979 at 2012-05-03 1840:
At present, most of the theoretical peak is calculated according to X4, because the CPU core can be a beat to execute two instructions, each instruction can perform two operations, so it is X4.
Intel's new architecture processor adds a new instruction set AVX, so one instruction can perform four operations, so the processor for the new schema ...
Excuse me, e7-8837 is x4 or x8.
zouming1979 (Contact TA in station)
e7-8837 I looked at it. AVX instruction set is not supported so it is X4
Bluesxn (Contact TA in station)
5/F: Originally Posted by zouming1979 at 2012-05-04 1300
e7-8837 I looked at it. AVX instruction set is not supported so it is X4

Here is a description of the number of floating-point operations per second:

At present, most of the theoretical peak is calculated according to X4, because the CPU core can be a beat to execute two instructions, each instruction can perform two operations, so it is X4.
Intel's new architecture processor adds a new instruction set AVX, so one instruction can perform four operations, so the processor for the new schema.

AVX Instruction Set The new instruction set of the Sandy Bridge and the Larrabee architecture Intel's microarchitecture has also entered a period of full speed, with Intel releasing its roadmap for 2010 years at the IDF summit, which ended in April 2010. Intel released its new processor microarchitecture, Sandy Bridge, in January 2011, and the new addition of instruction sets will also lead to increased CPU performance. Intel Company will bring a brand new instruction extension to Sandy Bridge Intel Advanced Vector Extensions (Intel AVX). The AVX is in the previous 128bit extension to and 256bit of SIMD (single instruction, multiple Data).    While the data transfer of Sandy Bridge's SIMD unit extends to 256bits, it is also improved, so the CPU kernel floating-point operation performance can be increased to twice times theoretically.    Twice times as close!!!!!!!!!!. Sandy Bridge is also said to be a floating point 8 times Ah pro ...