NVIDIA

NVIDIA Success Story



Life Sciences
 

The Max Planck Society for the Advancement of Science is an independent, not-for-profit research organisation based primarily in Germany. Like most of the world’s scientific research groups, the Göttingen Institute relies on computation as a vital tool for conducting and analysing the results of its work. For Professor Holger Stark and his group, their investigations into 3D Electron Cryomicroscopy have been significantly accelerated by the Institute’s adoption of parallel computing on the graphics processing unit (GPU) using NVIDIA’s CUDA and Tesla technology.

Challenge


Professor Stark’s work seeks to increase our understanding of the structure and 3D movement of tiny nanomolecular structures called macromolecules. Present in every living cell, these biological ‘machines’ are responsible for the most fundamental processes of life so gaining an accurate picture of their mechanisms is vital. For example, antibiotics work by acting on the functions of a particular kind of macromolecule in bacteria called a ribosome. A detailed understanding of these ribosomes and their functions is therefore very important in allowing medical researchers to develop effective drugs.

To produce detailed 3D images of macromolecules, Professor Stark’s team uses an electron microscope. Although modern electron microscopes are capable of resolutions higher than the distance between individual atoms, the biological structures being studied here would be destroyed by such an intense electron beam. To avoid harming the biological structures, the team cools their samples to a very low temperature and then uses a relatively low electron dose while observing the macromolecules’ structure and 3D movement. But since the low resolution produces “noisy” images which then need to be cleaned up, the researchers developed specialized 3D image processing tools that reduce noise and quickly align multiple images to improve accuracy.

Using a CPU cluster with 48 cores, the team was able to align 15,000 images in around seven days. However, at this rate, their target of aligning one million images would take 1.3 years for every macromolecule studied.

Solution


In February 2008, the Institute’s Göttingen facility became the first NVIDIA® Tesla™ installation in the world, with 200 GPUs in server configuration. By using the NVIDIA CUDA™ programming language to run their algorithms on the Tesla server solutions, the researchers were able to take advantage of the massively parallel computing power of GPUs and perform their calculations at a much faster rate.

With this GPU configuration, the alignment of one million images now takes 14 hours – over 800 times faster than what was possible on the previous CPU cluster. Due to these results, the Institute plans to expand its Tesla facility, allowing the image alignment process to be completed in just nine hours; placing the Göttingen GPU cluster’s theoretical performance on a par with the world’s most powerful supercomputers.

“NVIDIA’s GPU technology is exactly what our team needed,” says Stark, “and it came along at just the right time. The difference it makes to our work is fundamental – we can now perform calculations in just a few hours which were impossibly slow on our previous CPU-based solution. This technology is allowing us to accelerate innovation and discovery.”

Impact


New insights into the relationship between antibiotics and macromolecules in bacteria will lead to the development of more effective drugs and antibiotics. In time it is hoped that this will contribute to quicker recovery from illnesses and surgical procedures and ultimately saving lives.