Personal Computer Solves Complex Problems Tens of Times Faster Than Supercomputers?


Scientists from Skobeltsyn Institute of Nuclear Physics of the Lomonosov Moscow State University have come to the conclusion that a personal computer can be used for calculations of equations of a complex nature from quantum mechanics, which are usually solved with the use of supercomputers. Such a personal computer can get the job done much faster.

Vladimir Pomerantcev and Olga Rubtsova, senior researchers working under the guidance of Prof Vladimir Kukulin had been able to make use of an ordinary desktop personal computer with GPU to solve integral equations of a complex variety from quantum mechanics, which were previously solved by much more powerful and expensive supercomputers. According to Vladimir Kukulin, the personal computer can get the job done much faster; to put it into perspective, it requires only 15 minutes to get the work done which normally requires 2 to 3 days for the supercomputer to complete.

Equations, formulated in the 1960s by Russian mathematician Ludwig Faddeev, were used in the experiment. These equations describe the scattering patterns of a few quantum particles. As a result of these equations an entirely new field of quantum mechanics aka physics of few body systems appeared after this.

This area poses a great deal of interest to the scientists that are normally engaged in quantum mechanics, nuclear and atomic physics, as well as the theory of scattering. For a few decades, most notably after the pioneering work from Faddeev, one of the main purposes was to solve such complex equations. Due to the incredible complexity of these calculations, in conjunction with realistic interactions between a system particles made the reach of the researchers very limited; that is until the rise of supercomputers.

The situation did change, particularly after the SINP group made use of the Nvidia GPUs, which are normally found in game consoles, in their PCs. Vladimir Kukulin didn’t even use the most expensive processors for this “supercomputer” PCs, but used the ones that can currently be bought on the market for $300-$500.

The main problem in solving such scattering equations of multiple quantum particles was in the calculation of the integral kernel. This was a gigantic two-dimensional table, which consists of tens of thousands of rows and columns, with each and every element making a huge matrix, which results in extreme complex calculations. However, this table looked like a screen filled with billions of pixels and with a good GPU it was quite possible to calculate all these complex equations. By making use of the software developed by Nvidia along with their own programs, the researchers have been able to spread up the calculations through thousands of different streams and successfully solve the problem.

Vladimir Kukulin mentions that they have been able to reach a speed that was unsurpassed, and something to be dreamt of. Their program computes 260 million complex double integrals on a desktop computer in just three seconds flat. This is in no way comparable to the supercomputers. Even his colleagues in the University of Bochum in Germany can only dream of such computational power; for comparison Germans supercomputers based on expensive “Blue Gene” architecture can complete such task in two to three days, but Kukulin’s PC can do it merely 15 minutes without the need to spend any money.

The amazing thing from this research is that the desired quality of the graphics processors and a huge amount of software had been in circulation for over 10 years, but nobody made use of them for complex calculations, preference has always been given to supercomputers. The physicists from this university have surpassed and surprised their Western counterparts as well with such amazing results.

Vladimir Kukulin mentions that this opens up new avenues for them to analyze the nuclear and resonance chemical reactions. It could also be very helpful to solve large computing tasks in the field of plasma physics, geophysics, medicine, electrodynamics and various other areas of science. They are very eager to organize a training course for other researchers from various scientific streams that are unable to afford supercomputers, but could learn how to make their own personal computer handle what could usually be addressed as “supercomputational tasks”.

The complete results of this work have been published in the journal Computer Physics Communications.

  • Aj Otto

    Hmm, there should be other practical applications of 3d cards where Cpus are used because ‘you’re supposed to compute with a cpu’