The specialized climate model, which is built by researchers at Colorado State University, splits the globe into 20 million cells–one for each processor–and will be able to track the movement of storm systems and weather fronts. “Clouds are the most poorly simulated parts of the climate system,” says Michael Wehner, a climate researcher at the Berkeley lab. “And that has a lot of consequences.”
Currently, climate models estimate the behavior of clouds at a resolution of hundreds of kilometers. This means that similar climate models produce drastically different results. What’s more, higher-resolution models are important when making policy decisions at the regional level, Wehner says. For instance, governments can decide whether or not to move crops, budget for more water during potential future droughts, or improve infrastructure in coastal cities. “Improved models should make those decisions easier to justify,” Wehner says.
Still, it’s unclear whether the Berkeley researchers’ supercomputer design will actually save money in the long run, as the design and manufacturing costs can run to millions of dollars. And as the researchers dive deeper into testing the design, there could be challenges with designing the inter-chip communication system. “The problem when you put many of [the cores] together is that there are often inefficiencies in aggregating them,” says Mark Horowitz, a professor of electrical engineering and computer science at Stanford University, in Palo Alto, CA. “It’s a really interesting idea,” he says, “but it’s not clear whether it is going to be successful or not.”
Smaller design teams can now prototype and deploy faster.