jueves, 10 de mayo de 2012


In computer terms a benchmark is an application designed to measure the performance of a computer or any element thereof. For this purpose, the machine undergoes a series of workloads or stimuli of different type with the intention of measuring its response to them. This can be estimated under what tasks or stimuli a given computer behaves in a reliable and effective or otherwise shown inefficient. This information is very useful when selecting a machine to perform specific tasks in the post process and creation of audiovisual products, choosing the most appropriate for a given process.

The benchmark is also useful to estimate the level of obsolescence of a system or what technical performance can be improved through updates.
On the other hand, the benchmark can provide us all the technical specifications along with a computer's performance to different stimuli allowing for comparisons between different systems in line with their technical specifications and performance. The comparisons are useful in determining which technical characteristics are ideal for optimal performance in a specific task. A comparison between multiple computers from different manufacturers (with different specifications) allows us to determine a priori which are more suitable for certain applications and which ones are better for others.

Early in the 90's definition requires a supercomputer to produce comparable statistics. After experimenting with various metrics based on the number of processors in 1992, the idea of ​​using a list of production systems as a basis for comparison, at the University of Mannheim.

A year later, Jack Dongarra joins the project with the Linpack benchmark. In May 2003, establishing the first trial based on data published on the Internet:
  • Statistics Series supercomputers at the University of Mannheim (1986-1992)
  • List of sites world's most powerful computer, maintained by Gunter Ahrendt
  • Lots of information gathered by David Kahaner
To measure the power of systems using the HPL benchmark, a portable version of the Linpack benchmark for distributed memory computers.
Note that the list is not based on GRID computing systems or the MDGRAPE-3 supercomputer, which reaches a Petaflop being more powerful than any of the systems included in the list, you can not run benchmarking software used to not be a general-purpose supercomputer.
All lists published since the beginning of the project are posted on the website of the project, so it makes no sense to copy that information elsewhere.

Machines that have occupied the number 1

K_computer Fujitsu (Japan, June 2011 - present)
NUDT Tianhe-1A (China, November 2010 - June 2011)
Cray Jaguar (USA, November 2009 - November 2010)
IBM RoadRunner (USA, June 2008 - November 2009)
IBM Blue Gene / L (U.S., November 2004 - June 2008)
NEC Earth Simulator (Japan, June 2002 - November 2004)

1 comentario: