Maxeler Technologies delivers new pilot system on Quantum Chromodynamics (QCD) computations

11 April 2018

Maxeler Technologies, members of the CNIE Advisory Board, recently delivered a pilot system that has the potential of facilitating a breakthrough in energy efficiency of large-scale Quantum Chromodynamics (QCD) computations.

QCD is the theory for the strong force that binds together the fundamental particles, called quarks, to form protons and neutrons, as well as other hadrons. The actual size of quarks is not known, but measurements indicate that they are more than 1,000 times smaller than the proton.

Atomic and subatomic length scales. QCD is of particular relevance to subnuclear length scales.

"One of the Grand Challenges in Computational Physics, is calculating the binding of quarks by applying Monte Carlo methods to QCD. It is very pleasing to see that Maxeler is pushing the limits of what can be computed and bringing in a new era in large data computation", says Jerome Friedman who shared the Nobel Prize in 1990 for the discovery of quarks.

"The new machine outperforms existing QCD systems significantly in terms of power efficiency and computational density", according to Professor Georgi Gaydadjiev, Director of Maxeler IoT-Labs. The technology can now be used to build specialized data processing for QCD experiments and other applications.

"Maxeler's impressive QCD implementation is just one in the series of high impact applications across the physical sciences from experimental high energy physics to theory of materials, which have recently been migrated to Maxeler's Multiscale Dataflow technology", says Dr Vitali Averbukh, Department of Physics, Imperial College London.

The pilot system along with novel development tools for implementing large-scale compute-intensive scientific applications on Maxeler's proven Multiscale Dataflow technology was delivered under a contract awarded within a pre-commercial procurement of the EU PRACE-3IP project. Deploying this pilot system enables QCD physicists and other scientists to exploit this revolutionary technology, for instance to extend our understanding of how quarks bind to form matter and, ultimately, how our universe works.

Simplified loop flow graph of the BQCD, showing various compute stages and the amount of data (in GB) flowing between them during one "cycle" around this diagram

About Maxeler Technologies
Maxeler Technologies provides Maximum Performance Density for mission critical data processing applications and datasets beyond the currently manageable limits. Maxeler Dataflow Engines (DFEs) have been demonstrated to achieve faster, smaller, and smarter data processing in the datacenter as well as at the Edge of IoT. Ultimately, Maxeler Technologies provides a convenient way to extract value from ultra large datasets and complex data processing challenges.

Visit Maxeler’s website: http://maxeler.com/

Aerial view of the application bitstream as fitted to the reconfigurable device of MAX5. By zooming into this image one can distinguish between utilisation of logic, DSP or on-chip memory (FMEM) resources on the chip, which are colour-filled according to which kernel uses the resource. Note that the design contains many (minor) kernels not listed in the legend.