OREANDA-NEWS. Thanks to our Tesla Accelerated Computing Platform, researchers can now run computations and visualize the results on the same GPUs in a single system — either simultaneously (“in-situ”) or separately.

And Titan, the U.S.’s most powerful supercomputer, now holds the title of largest GPU-accelerated visualization system in the world.

That’s a major step forward. Researchers worldwide run high-powered supercomputer simulations to understand how the galaxy formed, how the human brain works and more. Then they create visualizations of their work to gain new insights.

Until recently, researchers did this work on two different systems. They used one for computation (running the science application). Then they switched to another to render the results visually. It could take days or weeks to complete the simulation before they could visualize it.

And depending upon the configuration of the data center, this could  require moving massive amounts of data from the simulation system to the visualization system. During this process, if researchers found an error or a parameter changed, they would have to re-run the entire simulation and render again. It’s an arduous process that could delay scientific results.

With 18,688 of our Tesla GPU accelerators, Titan at the Department of Energy’s Oak Ridge National Laboratory has been powering breakthroughs across a broad range of scientific domains since 2012.

Now, with an early implementation of hardware-accelerated in-situ visualization capabilities in Kitware’s ParaView application, Titan researchers can speed up their work by running real-time, interactive visualizations on the same GPUs that accelerated their simulations.

With Titan, they have access to thousands of GPU-accelerated nodes, versus hundreds on most visualization clusters. The enabling of hardware-accelerated graphics greatly increases the visualization capacity at the Oak Ridge Leadership Computing Facility.

A research team from the U.K.’s Imperial College is the first to jump on this opportunity.

They’re using GPU-accelerated PyFR simulations to study detailed profiles of air flow behind jet engines to better understand their acoustic properties. Airports have strict noise ordinances. So the team hopes to identify new ways to make engines quieter, such as altering the enclosure design.

The traditional two-system process for simulation and visualization was slowing down their work.

But with the ability to do both tasks on Titan, they can make key decisions on the fly. They can even change parameters mid-stream if needed to fine tune their work. That means they can get final results potentially 50 times faster.

Titan researchers won’t be the only ones who can take advantage of GPU-accelerated visualizations.

At the SC15 supercomputing conference this week, leading HPC visualization application developer Kitware is releasing new versions of its ParaView (v.5.0) visualization software and VTK (v 7.0) visualization toolkit.

Upgrading the rendering infrastructure to Open GL 3.x from version 1.1, the two applications supercharge rendering performance on GPU systems – 10 to 20 times faster compared to the previous versions. The new performance gains will benefit large and repetitive rendering tasks and improve interactivity for remote users.

Kitware also worked with NVIDIA to integrate large data volume visualization capabilities with NVIDIA IndeX into ParaView, mainly to address performance challenges in analyzing large data interactively as often required in exploratory science.

And with support for new EGL drivers that offer graphics context management functionality, applications like ParaView and UIUC’s VMD are easier to deploy on Tesla GPU-accelerated HPC systems.