Rpeak

This article looks at floating point computing power. For resistance values ​​see electrical resistance .

Rpeak is a theoretical value for the peak performance of a calculating machine when processing floating point numbers . The number of floating point additions and multiplications (with full accuracy) that an arithmetic unit performs in a unit of time (usually the cycle time) is counted and extrapolated to a full second based on the clock frequency. This value is then given in floating point operations per second ( FLOPS ). As an example, an Intel Itanium 2 performs four floating point number operations per cycle. At a clock rate of 1.5 GHz, this results in a value for Rpeak of 6 GFLOPS. The value represents a theoretical upper limit of the floating point computing power, which cannot be achieved in practical application. The most powerful mainframe in the world (as of June 2012) achieves 20  Peta FLOPS (PFLOPS), i.e. 20,000,000,000,000,000 floating point operations in one second.

Rmax

In contrast to Rpeak, Rmax is not a theoretically extrapolated value, but a value measured with test software ( benchmarking software ) in real operation of a computer system. The value achieved depends on the type of calculations carried out in the test software. The specification Rmax is therefore only comparable if the same test software was used with identical parameterization in the cases to be compared. In the TOP500 list of the fastest computer systems in the world, Rmax is determined using the HPL benchmark , in which linear systems of equations are solved. The above-mentioned supercomputer with Rpeak = 20.1 PFLOPS achieves Rmax = 16.3 PFLOPS, in this case about 81% of the theoretically possible computing power.

Rpeak to Rmax

Since Rpeak indicates the theoretical upper limit, Rmax ≤ Rpeak always applies regardless of the benchmarking software used. The quotient Rpeak / Rmax is also referred to as the efficiency of the system and in practice decreases with the number of nodes.

Individual evidence

1. ↑ What is the Rpeak? . Archived from the original on October 19, 2013. Info: The archive link was automatically inserted and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. Retrieved October 18, 2013.  
2. ↑ HPL - A Portable Implementation of the High-Performance Linpack Benchmark for Distributed Memory Computers . Retrieved October 18, 2013.
3. ↑ What is the Rmax? . Archived from the original on October 19, 2013. Info: The archive link was automatically inserted and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. Retrieved October 18, 2013. 
4. ↑ Documentation of the ZIVcluster - performance of the cluster . Retrieved October 18, 2013.