Over at the HPC-CH blog, they’ve got an interview with Argonne’s Venkartram Vishwanath on how he’s dealing with some extremely large simulation datasets. They’ve found that the visualization isn’t just good for finding insight in their data, but also a great workaround for IO bottlenecks that arise from the huge filesizes.
Venkartram agrees that one challenge of next generation simulations is that I/O will not keep up with the growth rate of computing capability. In his group at Argonne they are now working on efficient infrastructure and software to reduce the amount of data being written to storage to perform analysis, as well as in-situ visualization while the simulation is progress. This will facilitate the transformation of the data into insight.
PhysOrg has more information about the spectactular remote visualization and volume rendering shown in the San Diego Supercomputing booth at SC09, which makes use of Eureka, the TeraGrid, 200 NVidia GPU’s, and high speed networks to stream high-resolution images around the world.
“As a team, we were able to link institutions across the country and leverage high performance computing, visualization resources, high speed networks and advanced displays in real-time,” said Joe Insley, principal software developer at Argonne. “But what was really wonderful was seeing the scientists get excited about the possibilities that this will enable.”
The simulation was part of a 2009 TeraGrid allocation using a 4000^3 grid of 64 billion dark matter particles, running for over 4 million CPU hours. With data that massive, remote visualization isn’t just a nicety, it’s a requirement.
The San Diego Supercomputer Center drew a standing-room only crowd when they fired up their connection to the Eureka visualization cluster at Argonne to stream the massive visualization of the universe to the show floor.
This visualization of the Universe as it condenses around fluctuations in the density of dark and ordinary matter is a result from a collaboration between Argonne National Laboratory and the San Diego Supercomputer Center/University of California, San Diego. In a demonstration at SC09 visualizations of a 4096^3 data volume was streamed from Eureka, the graphics cluster at Argonne, over ESnet, filing a 10Gb/s network link, to an OptiPortal in the SDSC booth on the exhibit floor.
You can read an audience account of the event at InsideHPC, and technical details at the Argonne website.
Argonne National Labs, part of the US Department of Energy, is holding an open house later this month, August 29th from 9am – 4:30pm, which will include tours of some of the major labs and a chance to see some massive visualizations.
Nearly 100 engaging exhibits, demonstrations, tours and presentations are being planned for a day of fun and education for the entire family.
A program of attractions and events, including a map of the Argonne campus, will be available free at the Argonne gates and is downloadable here.
As a followup to PhysOrg’s Argonne announcement of the massively parallel simulation and visualization of the supernova explosion, New Scientist has gathered some of the images for a fantastic look at how the data really look.
Over at Argonne, researchers are setting new records for interactive visualization of large datasets with volume-renderings of 80-billion voxels from an astrophysics dataset. What’s particularly surprising is that the entire process is being run distributed in parallel on the IBM Blue/Gene, without GPU Accelerators.
Argonne researchers wanted to know if they could improve performance by skipping the transfer to the GPUs and instead performing the visualizations right there on the supercomputer. They tested the technique on a set of astrophysics data and found that they could indeed increase the efficiency of the operation.
“We were able to scale up to large problem sizes of over 80 billion voxels per time step and generated images up to 16 megapixels,” said Tom Peterka, a postdoctoral appointee in Argonne’s Mathematics and Computer Science Division.
At the recent SciDAC conference in San Diego, 10 videos won awards at the Electronic Visualization and Poster Night. One of those awards went to Argonne for their “Turbulent Flow of Coolant in an Advanced Nuclear Reactor”.
Both the visualizations and the computer runs for the winning entries were done at the Argonne Leadership Computing Facility (ALCF). The computations were carried out on one of the world’s fastest and most energy-efficient supercomputers: Intrepid, Argonne’s IBM Blue Gene/P. The visualizations were performed on Eureka with software developed at Argonne. Eureka is also located at the ALCF and is one of the world’s largest graphics processing units, providing more than 111 teraflops and over 3.2 terabytes of RAM.
The Department of Energy (DOE) is putting 1.3Billion Processor Hours out to the public for simulation and research of several phenomena. The hours are split between the Oak Ridge Cray XT “Jaguar” and Argonne IBM BlueGene/P “Intrepid” systems, and are for unclassified open research.
In 2009, 900 million processor hours were up for grabs (a million processing hours would take 1,000 processors 1,000 hours, or around 41 days), but both computers received huge performance boosts this year. Jaguar’s processor count has shot up from 31,328 to 180,832, while Intrepid now boasts 163,840 from 32,768. Jaguar’s peak performance is now a blistering 1.64 petaflops (a quadrillion and a half floating point operations per second), making it the second most powerful supercomputer on Earth.
While the article talks alot about the science the new machines facilitate, I see nothing about how they intend to perform Analysis or Visualization of this scale of data. An oversight? I know some DOE folks are watching, care to elaborate?
The ALCF’s ability to visualize such enormous quantities of data is made possible by of the world’s largest graphics processing units (GPU). Nicknamed Eureka, this installation of NVIDIA Quadro Plex S4 external GPUs allows researchers to better understand the data they produce with Intrepid at the ALCF. The powerful installation provides more than 111 teraflops and more than 3.2 terabytes of RAM.
Eureka has some interesting hardware features including a special configuration of Quadro Plex S4′s, a nine-switch 2048-connection network interconnected, and a special bank of 10,000 disk drives all packed into 4 racks.
Over at the HPC-CH blog, they’ve got an interview with Argonne’s Venkartram Vishwanath on how he’s dealing with some extremely large simulation datasets. They’ve found that the visualization isn’t just good for finding insight in their data, but also a great workaround for IO bottlenecks that arise from the huge filesizes.
PhysOrg has more information about the spectactular remote visualization and volume rendering shown in the 
The San Diego Supercomputer Center drew a standing-room only crowd when they fired up their connection to the Eureka visualization cluster at Argonne to stream the massive visualization of the universe to the show floor.
As a followup to 
Over at Argonne, researchers are setting new records for interactive visualization of large datasets with volume-renderings of 80-billion voxels from an astrophysics dataset. What’s particularly surprising is that the entire process is being run distributed in parallel on the IBM Blue/Gene, without GPU Accelerators.
The Department of Energy (DOE) is putting 1.3Billion Processor Hours out to the public for simulation and research of several phenomena. The hours are split between the Oak Ridge Cray XT “Jaguar” and Argonne IBM BlueGene/P “Intrepid” systems, and are for unclassified open research.
VizWorld.com 
Comments