ScREC

ScREC is a supercomputer developed by the Research Centre for Modeling and Simulation (RCMS) at the National University of Sciences and Technology, Pakistan (NUST) in Islamabad, Pakistan. With a 132 teraflops performance, it is currently the fastest supercomputer in Pakistan.[1]

ScREC
ActiveSeptember 2012[1]
OperatorsResearch Centre for Modeling and Simulation (RCMS), NUST
LocationResearch Centre for Modeling and Simulation, NUST, Islamabad, Pakistan
Storage21.6 TeraByte[1]
Speed132 TeraFLOPS[1]
PurposeMultipurpose[1]
Web site

System specifications

ScREC is able to perform parallel computing and has a performance speed of 132 teraFLOPS (trillion operations per second). It is the fastest running graphics processing unit (GPU) parallel computing system to have been developed in Pakistan.[1] The supercomputer has multi-core processors and graphics co-processors, with an inter-process communication speed of 40 gigabits per second. According to system specifications, the computer cluster consists of 66 nodes equipped with 30,992 processor cores. Additional component details include:

  • 32 dual quad core computer nodes (256 processor cores)
  • 32 Nvidia graphics processing units
  • QDR InfiniBand interconnection
  • 21.6 TB storage
gollark: The second one is less controversially "yours" than the first.
gollark: Those are different things, though. A face recognition model is going to be trained on a lot of people's faces, and can then generically match faces together. You can then use that to encode someone's face into an embedding vector you can use for matching.
gollark: I had assumed this stuff was now ML-based and so you would just compare embedding vectors or something.
gollark: What are they eigenvectors *of*, exactly?
gollark: eigen is "own" or something, and apparently people prefer that over "characteristic vector/value".

See also

References
  1. "Fastest supercomputer is out". Daily Times (Pakistan). 19 September 2012. Retrieved 7 October 2012.
This article is issued from Wikipedia. The text is licensed under Creative Commons - Attribution - Sharealike. Additional terms may apply for the media files.