Injection molding in automotive design
Injection-molding simulations are invaluable to car makers, Autodesk's Martin says. Injection molding is a process for producing parts from plastic materials. Simulations show whether an injection mold - such as a bumper, for instance - will cause denting and how the mold will fit with other parts of the car. They also reveal any defects. Designers consider many variables: the temperature of the mold, its geometric shape and how the injection-mold process will work with certain materials.
A single physical prototype of a fender can cost more than $1 million, Martin explains, so the better the simulation, the fewer prototypes that have to be built and the lower the production costs.
Simulations "used to require a significant cluster-computing installation, but we are achieving the same level of power with current desktop computers," says Martin, who says the desktop advances that played the biggest roles in making that possible were the move to multicore processing, the use of multiple GPUs and 64-bit throughput. Martin uses standard desktop computers that can be purchased at Wal-Mart, with the latest 3D-capable GPUs and Intel dual-core CPUs.
The automotive industry is also using desktop software to model car designs. With such tools, the manufacturers can create complex renderings that rival - or even outshine - the graphics in the best video games. Interestingly, when these models are created on desktop PCs, they are often used in marketing literature and in TV commercials. Martin says it was possible to take the models created on supercomputers and transport them to PCs, but today's PC-to-PC file transfers are simpler because the file formats are compatible across applications and the transfers take place over a standard network, making it possible for the images to reach a greater number of users.
Here's an example of modeling on the desktop: Designers often create car-paint models, examining the metallic flakes and how they look on certain plastics, or whether the paint looks dull in certain lighting conditions. The more accurate the models, the more process-intensive they are. In the past, an HPC environment was required because of the high number of polymers used in car paint -- around 8,000. Martin says modern desktops can now handle this kind of high-speed processing.
Web-based computational search
One of the most interesting ways HPC is coming to the desktop is through the Web. The best example of this right now is on WolframAlpha, a "computational knowledge engine" on the Web that's designed to, in the site's own words, "collect and curate all objective data... and make it possible to compute whatever can be computed about anything."
The sea change here is that, while the searches you launch at WolframAlpha.com are still conducted on a supercomputer, the results return almost instantly, right in your browser. For example, when you type the simple word moon, you trigger a complex calculation for the moon's orbit relative to the Earth, along with average distances over a historical period.
Schoeller Porter, an architect at Wolfram Research Inc., says the calculations can occur in near-real-time because of the decreased cost of HPC components. In the past, this kind of complex calculation would require setting up a batch request for a supercomputer that might take a few minutes or hours, or even a day. That wasn't necessarily because the calculations themselves take a long time, but because the processing resources were so expensive.
To test the yourself, type these calculation requests - asking about the tides in Honolulu in six months - into the WolframAlpha search engine:
* International space station jun 24
* skychart Timbuktu yesterday at 8:00pm
* NACA 2014 15 degrees
* tides in honolulu in 6 months
* y'' + sin y = x
When you run searches on WolframAlpha, you can change any variable at will from your desktop. You'll notice that some of the queries take longer than others, depending on how long it takes the supercomputer to determine the result.
While Google searches also rely on a supercomputer or cluster on the back end, they do not involve the same computational algorithms as WolframAlpha. Google searches for information and provides relevant links; WolframAlpha feeds the user actual information, not links, and that information may be pulled from a knowledge base or generated by calculations it performs.