A cluster of standard Apple laptops could work as a supercomputer, enabling people to do computationally demanding tasks in a cheaper way.
Researchers who need powerful computers to run their experiments have been moving away from monolithic supercomputers, which are expensive to build and maintain. Instead, they have often preferred clusters of devices originally designed as graphics cards for desktop computers. These graphics processing units (GPUs) can perform large numbers of parallel calculations, but strong demand for them, including from cryptocurrency miners, has pushed prices ever higher.
Now, and Connor Kenyon at the University of Massachusetts, Dartmouth, have found that Apple laptops may work as an alternative.
Advertisement
Apple’s latest laptops, desktops and iPads contain a range of chips called M1, which are rooted in technology from UK-based processor manufacturer Arm. Each chip integrates almost all the components needed to create a computer, rather than having different functions spread across numerous chips.
Capano compared Apple’s M1 and M1 Ultra chips with state-of-the-art GPUs, including several versions of the NVIDIA V100, using a tool called the benchmark, which measures various facets of performance. Facebook’s parent company Meta is using similar NVIDIA chips to build the world’s most powerful AI-specific supercomputer.
Both Apple chips beat three versions of the NVIDIA GPU in three key areas of the benchmark. And, even though the NVIDIA chips can be purchased as individual components whereas you need to buy Apple computers containing M1 chips rather than being able to get the components standalone, the cost of getting hold of the M1 chips was still competitive.
At the time Capano and Kenyon’s paper was written, the cheapest computer with an M1 chip was about $900 and the cheapest computer with an M1 Ultra chip was $5000. The cost of the NVIDIA GPUs ranged from about $3000 to $11,000. The UK’s most powerful supercomputer, ARCHER2, cost around £79 million.
There is precedent for stacking up consumer electronics to construct supercomputers. In 2010, the 33rd most powerful supercomputer in the world was actually an array of . Capano says the research and development required to create tailored chips means that there will always be a role for consumer devices. “You’ve got to make do with what you can,” he says.
There would be hurdles to repurposing the M1 devices, though. They use proprietary technology, so a method would be needed to efficiently tie them together and split a large computational problem across them.
“This means having to write code specifically for these processors, which is kind of a pain,” says Capano. For the gravitational wave astronomy that his team wants to do, it will involve writing code in , which is used to communicate with graphics hardware. Apple didn’t respond to a request for comment.
Reference:
