Snow Falling On Silicon: A Norwegian Uses GPUs to Simulate Flurries

by Tony Kontzer

Fact is, computer scientists sometimes just like to play with technology.

Such was Anne Elster’s motivation when she resolved several years ago to start working on real-time snow simulations.

It’s probably important to point out that Elster, who was talking about her work at the GPU Technology Conference, is an associate professor at the Norwegian University of Science and Technology.  Rumor has it Norway gets a bit of snow.

Elster thought focusing on snow would be a fun project for her students, allowing them to combine graphics processing with powerful simulations. She also figured it would be a nice tool for recruiting more post-graduate students.

The precursor to the project was an effort a decade ago by Elster and her students to run real-time simulations of the behavior of smoke using a dual-core laptop. Their goals at the time seem tame today: 20 frames a second, about a third of the current standard for real-time simulations.

Fast forward to 2007, and Elster started the snow-simulation effort by parallelizing on a multi-core laptop. But she eventually saw that GPUs would be a better choice for providing the required computing power.

The following year, she and her students were able to simulate several million snow flakes as particles, complete with wind field interactions, all made possible by GPUs.

Elster and her students have since expanded their simulations to work on including features such as avalanches, geysers and waterfalls. Now, they’re working on incorporating more realistic terrains, and are considering adding ray tracing to simulate how snow interacts with light.