Open-source GPU technology for supercomputers

PICTURE: Vladimir Stegailov, professor at HSE University After

Credit: Vladimir Stegailov

Researchers at the HSE International Laboratory for Atomistic Modeling and Multiscale Analysis at HSE, JIHT RAS and MIPT compared the performance of popular molecular modeling programs on GPU accelerators produced by AMD and Nvidia. In an article published by the International review of high performance computing applications, researchers ported LAMMPS to the new open-source GPU technology, AMD HIP, for the first time.

Researchers took an in-depth look at the performance of three molecular modeling programs – LAMMPS, Gromacs, and OpenMM – on Nvidia and AMD GPU accelerators with comparable peak parameters. For the tests, they used the model of ApoA1 (Apolipoprotein A1) – apolipoprotein in blood plasma, the main protein that carries “good cholesterol”. They found that the performance of search calculations is influenced not only by hardware parameters, but also by the software environment. It has been found that inefficient performance of AMD drivers under complicated compute kernel parallel launch scenarios can cause significant delays. Open source solutions always have their drawbacks.

In the recently published article, researchers were the first to port LAMMPS to a new open-source GPU technology, AMD HIP. This developing technology looks very promising as it allows efficient use of code both on Nvidia accelerators and on new AMD GPUs. The developed LAMMPS modification has been released in open source and is available in the official repository: users all over the world can use it to speed up their calculations.

“We have thoroughly analyzed and compared the GPU accelerator memory subsystems of the Nvidia Volta and AMD Vega20 architectures. I found a difference in the parallel launch logic of GPU cores and demonstrated it by viewing the program profiles. The memory bandwidth and latencies of the different levels of GPU memory hierarchy as well as the efficient parallel execution of GPU cores – all of these aspects have a major impact on the actual performance of GPU programs, ” said Vsevolod Nikolskiy, PhD student at HSE University. and one of the authors of the article.

The authors of the article argue that the participation in the technology race of contemporary microelectronics giants demonstrates a clear trend towards a greater variety of GPU acceleration technologies.

“On the one hand, this fact is positive for end users, as it stimulates competition, increasing efficiency and lowering the cost of supercomputers. On the other hand, it will be even more difficult to develop effective programs due to the need to take into account the availability of several types of GPU architectures and programming technologies ”, commented Vladimir Stegailov, professor at the HSE University. “Even supporting program portability for regular processors on different architectures (x86, Arm, POWER) is often complicated. The portability of programs between different GPU platforms is a much more complicated issue. The open-source paradigm removes many barriers and helps software developers of large and complicated supercomputers.

In 2020, the graphics accelerator market experienced a growing deficit. Popular areas of their use are well known: cryptocurrency mining and machine learning tasks. At the same time, scientific research also requires GPU accelerators for the mathematical modeling of new biological materials and molecules.

“Creating powerful supercomputers and developing fast and efficient programs is how the tools are prepared to solve the most complex global challenges, such as the COVID-19 pandemic. Computational tools for molecular modeling are now used around the world to research ways to combat the virus, ”said Nikolay Kondratyuk, researcher at HSE University and one of the authors of the article.

The most important programs for mathematical modeling are developed by international teams and researchers from dozens of establishments. The development is carried out in the open-source paradigm and under free licenses. Competition from two contemporary microelectronics giants, Nvidia and AMD, has led to the emergence of a new open-source infrastructure for programming GPU accelerators, AMD ROCm. The open-source nature of this platform gives hope for maximum portability of the codes developed with its use, to supercomputers of various types. Such an AMD strategy is different from the approach of Nvidia, whose CUDA technology is a closed standard.

It didn’t take long to see the response from the college community. Plans for the biggest new supercomputers based on AMD GPU accelerators are nearing completion. The Lumi in Finland with 0.5 performance exaFLOPS (which is similar to the performance of 1,500,000 laptops!) Is under construction. This year, a more powerful supercomputer, Frontier, is expected in the United States (1.5 exaFLOPS), and in 2023 – an even more powerful El Capitan (2 exaFLOPS) is expected.


Warning: AAAS and EurekAlert! are not responsible for the accuracy of any press releases posted on EurekAlert! by contributing institutions or for the use of any information via the EurekAlert system.

Source link

About Mariel Baker

Check Also

Kyoto University of Japan lost 77TB of important data due to Hewlett Packard Japan supercomputer-

Kyoto University of Japan yesterday issued a statement saying that from December 14-16, their supercomputer’s …