Slashdot Mirror

This is seriously useful for massive scientific visualization... where raw rendering speed isn't always the bottleneck (but of course, faster is better).

We do simulations on supercomputers that generate terabytes of output. You then have to use a smaller cluster (typically 1000-2000 processors) to read that data and generate visualizations of it (using software like Paraview ( http://www.paraview.org/ ) ). Those smaller clusters often don't have any graphics cards on the compute nodes at all... and we currently fall back to Mesa for rendering frames.

If you're interested in what some of these visualizations look like... here's a video of some of our stuff: https://www.youtube.com/watch?...

What other GPU enabled software runs on Windows, Mac, and Linux?

For starters: Every game that makes use of 3D and is available for the three platforms, scientific software like Paraview, Slicer 3D, 3D rendering software like Blender, the famous video player VLC, ...

I write massively parallel scientific simulation software for a living (the kind that runs on the biggest machines in the world)... and trying to come up with a way to display GBs or TBs of information from some of our largest simulations can be _tough_.

We use several open source packages ( Mostly http://www.paraview.org/paraview/resources/software.php and https://wci.llnl.gov/codes/visit/ ), but most of our best visualizations are actual done using a commercial package ( http://www.ceisoftware.com/ )

For some examples check out the YouTube video here: http://www.youtube.com/watch?v=V-2VfET8SNw

(That's me talking in the video). Those aren't necessarily our best visualizations - just some that happen to be on YouTube...

We find that the reactions to these simulations are mixed. They are certainly eye-pleasing... but sometimes if you go too far in making it look good it can actually turn scientists off. They will start to think that it looks "too good to be true" (I literally had a senior scientist in a room of 200 stand up at the end of one of my presentations and proclaim that "This is too good to be true!"). Because of this we try to do do just enough visualization that you can see all of the features of the simulation and understand what's happening without going overboard.

You have to realize that a lot of scientists still remember the days when they created line plots _by hand_ for publications! I suspect that as young scientists come up through the ranks this feeling that "slick graphics = not real" will go away.

At least, I hope....

> 3D data is all but ungraphable on Linux systems anyway,

? eh, you must be looking very hard in the wrong place.

for one, try Paraview.
for two, try VisIt.
for three, there are many more like that ... (including my favorite)

• start with a wide variety of platforms to begin with, and
• maintain coding standards that spell out common problems with compilers

nobody would be able to make a living writing software

I have managed contracts to fund developers working on open source software projects. My employer pays programmers to write software and to release it with an open source license. The Department of Energy (our funding source) has spent literally millions of dollars over the last few years on projects like this.

I contest the claim that writing open source software entails no monetary compensation to the software developer.

Sorry, but I fail to see what Blender and the GIMP have to do with real scientific visualization. Blender is for 3D modelling, and the GIMP is for image processing.

If you're looking for complete, open source scientific visualization and data analysis packages, try VisIt, which supports dozens of input formats and runs on Linux, Windows, and MacOSX. Pick it up at http://www.llnl.gov/visit, or get the latest binaries from FTP here.

I have less knowledge of ParaView, but it is also free: http://www.paraview.org.

Both of these are also developed in part by the national labs; they can run parallel to handle terabytes of data, so if you've got small dataset they should be smokin' fast, and if you've got your own cluster you should be able to visualize some huge data.

If you're looking for just a toolkit to build your own application, try OpenDX or VTK.

Once you have your visualization cluster, decided on the CPU, the interconnect, the OS, etc., you might ask what kind of application you can run on it.

If you take a look at ParaView, a project built on top of VTK by Kitware, you might notice that Kitware is funded by ASCI. And what is ASCI? It is a cooperative program spread across Sandia, Lawrence Livermore, and Los Alamos National Labs. The labs do indeed fund a good bit of open-source software.

It's not "false savings" in my case. We compile ParaView frequently during development. It's over 1200 C++ classes and takes at least an hour to compile from scratch on a fast machine. Even small changes to a single header file can cause large numbers of files to be recompiled. And we like to run continuous, experimental, and nightly builds (with DART) for testing purposes. Since we run ParaView on clusters, it's nice to be able to build in parallel, too. With the demise of pmake, it's good to see some open source tools for distributed compilation being developed. distcc isn't ideal (cpp and ld must be run on the head node and there can only be a single head that passes does load distribution), but it's a lot better than being stuck on a single node.