Reg comments9

Supercomputer helps boffins crack 3D material sims...

Tokyo Tech nabs Gordon Bell Prize and $10k for metal materials work

server room

SC11 So who cares about dendrites and dendrite solidification? If you’re an auto manufacturer facing a mandate to radically increase the fuel economy of your cars, you care a lot. Or if you’re working on new jet engines and need to cut some pounds while increasing durability. Actually, if you rely on any alloy that has to have those ‘just right’ properties, then dendrite solidification is crucially important to you.

When you look at a metal or alloy under a very strong microscope, you’ll eventually see that they’re all made up of tiny crystalline grains – which are, in turn, composed of tiny interlocked snowflake-like structures that scientists have dubbed ‘dendrites’. This structure comes about when the substance changes form from a liquid (like the liquid metal in a Terminator T-1000) to a solid.

The dendrites dictate the properties of the material – its hardness and flexibility, for example. Being able to examine the distribution of many multiple dendrites is important because it’s this distribution that shows if the new alloy is fit for the purpose for which it is being designed. While seeing dendrites is important, having the ability to simulate dendrites and dendrite distribution and then predict the effects of changing the alloy recipe is a very big deal... and that’s where petascale supercomputing comes in.

Dendrites simulated in
2D (click to enlarge)

The scientists at Tokyo Institute of Technology have been doing quite a bit of work in this area using their TSUBAME 2.0 supercomputer. This machine – a 2.4 petaflop collaboration between HP and NEC composed of 73,000 CPU cores and 57,000 NVIDIA Fermi GPU cores – is currently #6 on the Top500 list.

Before TSUBAME, scientists were limited to simulating dendrites in 2D (left), which was helpful, but of limited benefit.

With TSUBAME, they can throw 16,000 CPU cores and 4,000 GPU cores at the problem and get a 3D simulation of a bunch of dendrites (below right) and then a close-up view of a single dendrite (below left).

3D simulation of a bunch
of dendrites (click to enlarge)

Close-up view of a single
dendrite (click to enlarge)

With this capability they’ll be able to simulate more alloys to a much finer degree and, assumedly, fine-tune them.

For proving that they can use 2.0 petaflops on a practical scientific application, the Tokyo Institute of Technology researchers were awarded the Gordon Bell Prize for Special Achievement in Scalability and Time-to Solution for their paper titled Peta-scale Phase-Field Simulation for Dendritic Solidification on the TSUBAME 2.0 Supercomputer. Gordon Bell is also known as the Nobel Prize of supercomputing.

The paper is fairly deep on the math and science. Here’s a small sample ...

(Click to enlarge)

For more details on the project and the paper, click here. ®

Sign up to our Newsletter

Get IT in your inbox daily

Biting the hand that feeds IT © 1998–2017