For the first time, there are over 100 accelerated systems in the ranking of the 500 most powerful supercomputers in the world, for a total of 143 petaflops, therefore, more than a third of the total FLOPs. Among these, there are 70 supercomputers based on NVIDIA Tesla GPUs , and more specifically 23 of the 24 new systems entered the ranking, a number that confirms an annual growth that in the last five years has stood at around 50%.

There are three main reasons why accelerators are increasingly used in high performance computing:

•  Advances related to Moore’s Law continue to slow, forcing the market to make computational power available more efficiently.
• Hundreds of applications (including the majority of those commonly used by users) are GPU accelerated today.
• Furthermore, it should be considered that even modest investments in accelerators can now lead to significant increases in throughput, optimizing the efficiency of hyperscale datacenters and supercomputers.

“Someday all supercomputers will be accelerated.” Said Jen-Hsun Huang , co-founder and chief executive officer of NVIDIA. “Several facilities that house the world’s leading supercomputers have transitioned to GPU-accelerated computing, as the TOP500 reflects today. Considering the fact that research is progressing rapidly and that researchers are moving towards computational computing, machine learning and visualization, we expect this trend to be further consolidated. “

Numerous world-leading systems use NVIDIA Tesla accelerators , including the fastest supercomputers in 10 countries (United States with Titan, at Oak Ridge National Laboratory, and Russia with Lomonosov 2, at Moscow State University) plus the most powerful supercomputer present in Europe, the Swiss Piz Daint, at the Swiss National Computing Center.

Taking into account that the size of transistors is approaching the atomic scale, it becomes increasingly difficult to improve the performance of microchips without increasing the power or cost exponentially. Although the market can no longer rely on a doubling of performance every 18 months, computational needs continue to increase. This has led to a growing popularity of accelerators, which work together with CPUs to improve performance with technical and scientific applications.

Since 2008, the Tesla platform has grown steadily in the number of applications supported in science, for data analysis and much more, up to the current 370 applications supported by GPUs. A new study from Intersect360 Research, a technology research firm, shows that approximately 70% of the 50 most used HPC applications and 90% of the top 10 support GPU accelerated computing. These include the fluid dynamics application ANSYS Fluent, GROMACS for molecular dynamics and now also VASP, an atomistic simulation application used by researchers around the world to model the behavior of individual atoms electronically.

One of the study’s authors, Addison Snell , CEO of Intersect360 Research , commented,
“Accelerated computing has reached a tipping point in the HPC industry and sees NVIDIA Tesla GPUs as the market leader. The use of accelerators and the availability of GPU-accelerated code for major HPC applications are also steadily growing. “

Supercomputing datacenters and hyperscale datacenters can cost hundreds of millions of dollars. In the past, the expected constant increase of Moore’s Law allowed them to always update them with new CPUs to keep up with the ever-increasing needs, but this is no longer possible. With the advent of GPU-accelerated computing, these investments in data centers can be made by adding new NVIDIA Tesla accelerators, which increase throughput while meeting new needs.