The International Supercomputing Conference (ISC) in Hamburg, Germany, is a great place to connect with the European and Asian high performance computing (HPC) communities, and ISC’12 was a milestone event for NVIDIA.
It’s hard to overstate the impact Tesla GPUs and CUDA have had on the HPC market since their launch five years ago. In that time, GPU computing has opened the doors to a new wave of industrial and scientific discovery.
The Top500 list of the world’s most powerful supercomputers released at ISC’12 in June underscores just how far we have come.
|Tesla Fermi was a significant inflection point in
terms of number of GPU supercomputers in the
There are 52 systems powered by NVIDIA Tesla GPUs on the new list, which is a fourfold increase compared with just one year ago. When we launched our Fermi-based Tesla GPUs in November 2010, only 10 GPU-based systems made the list, which shows how the adoption of GPU technology has steadily marched on.
What is particularly interesting about the current list is that GPUs are bringing supercomputing power to increasing numbers of mainstream universities and research centers. By far the biggest GPU increase occurred for systems ranking between Nos. 101 to 500 – a mammoth 680 percent jump.
While a spot on the Top500 list used to be for privileged institutions with large budgets, the latest list highlights that GPU computing is enabling affordable supercomputing capabilities for just about any university or research institution. Formerly, researchers had a hard time getting access to a big CPU-only supercomputer to conduct their research. Now, each university department can have a small dedicated GPU cluster that delivers 5 to 10 times more performance.
A great example is the University of Bristol in the U.K., where researchers working with teams in Thailand recently made a breakthrough discovery about the deadly H1N1 virus. In 2009, H1N1 killed more than 500,000 people worldwide, largely because the virus mutated frequently, making once-effective drugs like Tamiflu ineffective.
Using a small cluster with just eight Tesla GPUs, the researchers ran complex computer simulations that for the first time revealed new ways in which inhibitor drugs can be quickly designed to address these mutations, and possibly reduce the deadly impact of future epidemics.
The H1N1 research project normally would have required months of computer time on a large CPU cluster, which simply wasn’t an option for the Bristol researchers. However, with the small GPU cluster, the team was able to uncover the new H1N1 data in half the time and using only one-fifth the number of servers.
Parallel Programming Education Takes Off
In addition to greater access to affordable, high-performance GPU accelerators, the widespread availability of parallel programming courses are helping researchers take advantages of parallel computing. The CUDA parallel programming model has given universities an easy way to teach parallel programming integrated into the most familiar programming languages of C, C++ and Fortran. With more than 580 universities teaching parallel programming with CUDA, the academic research community has widely embraced GPUs as the way to accelerate their race to better science.
I truly believe that the biggest impact of GPU computing is in democratizing supercomputing. NVIDIA Tesla GPUs not only power the fastest supercomputers in Brazil, China, India, Russia and Spain, but also power tens of thousands of small department clusters in universities all over the world.
The Top500 list is important, but the biggest impact of research will come from giving access to supercomputing to the next 1,000 research institutions.
[Featured image credit: Václav Pajkrt]