Publish at January 14 2025 Updated January 14 2025
Your computer has a number of processor cores with different specializations, cache memories and peripherals of all kinds. In a supercomputer, the physical infrastructure is not only several orders of magnitude more complex, but also undergoes frequent evolution and transformation. So being able to physically position the available hardware resources means determining the best resource allocations for optimum performance.
For a supercomputer like the Frontier, one of the world's most powerful calculators, capable of performing over 1 billion billion operations per second, every microsecond counts. Parallel calculations are the norm, and you need to be able to find the best working configuration.
"The hwlocsoftware , whose origins date back to Samuel Thibault's thesis, was created to facilitate the identification and location of each of a supercomputer's resources (processor cores, cache memory, peripherals, etc.).
hwloc is an open-source software package now used worldwide and made available by the Inria Centre at the University of Bordeaux since 2009. Without the mapping it enables, it would be trial and error trying to find the best way to organize calculations.
For the full article: Brice Goglin and Samuel Thibault: HPC cartographers
Domaine de Voluceau
Rocquencourt - B.P. 105
78153 Le Chesnay
France
Tél.: 33 (0)1 39 63 55 11
You might also like:
Sports performance and connected watches: extracting relevant information
The aim of the INRIA team is to use the easily accessible signals from connected objects to develop algorithms capable of interpreting them to extract useful information for athletes.
During the Cold War, the Americans were trying to develop a communications technology capable of linking computers across the country. Their work over several decades led to the creation of the Internet.
The wonderful world of intelligent algorithm bias control
The biases that algorithms can develop are known and numerous. While everyone recognizes the need for regulation, the challenge of drafting enforceable and effective regulations is on the table. The Regalia pilot project is trying to devise ways.
Graph-based learning for frugal artificial intelligence
Data modeled in graphs (which can be multi-dimensional) allow the detection of relationships that are practically invisible in other forms and, above all, with much less computational and energy resources.
No need to say what you think, I'll know.
"The realization of this new way of understanding and visualizing cognition offers a major opportunity: that of being able to guess, from an image of the brain, the cerebral activity at play."
