Microsoft foresees a new type of AI PC: A Surface designed with help from machines

For now, Redmond is dogfooding Azure for product simulations

Microsoft has bragged that its own Azure HPC service was able to reduce the length of its Surface laptop design process – most notably for a hinge, which was reduced to one iteration, and hopes to use AI to do even better in future.

According to principal engineer Prasad Raghavendra, Abaqus FEA software has been implemented into Azure HPC since 2015. By 2016, Redmond had fully migrated product level structural simulations for Surface Pro 4 and the original Surface laptop to Azure HPC from on-premises servers.

For those not versed in the mechanical design world, it works like this: computer-aided design (CAD) models – or digital drawings of a laptop complete with all its components – are translated into finite element analysis (FEA) models. The FEA models can then simulate things like the effects of temperature, or the forces experienced when a machine is dropped. That informs any adjustments or design choices that need to be made before a physical prototype is manufactured and run through real-world tests.

"In a few days, hundreds of simulations are executed to evaluate various design ideas and solutions to make the device robust," explained Raghavendra.

In the case of the aforementioned hinge, a graphic depicting its movement when a laptop is dropped and lands on a corner – as laptops tend to fall – allowed the engineering team to visualize the impact and stress levels experienced by its inner parts.

That dynamic drop simulation was executed on hundreds of cores of an Azure HPC cluster using Abaqus Explicit solver – the simulation tool used for brief transient and dynamic events like dropping heavy electronics or car crashes. In this case, the solvers are optimized for Azure HPC clusters specifically, allowing the simulation to scale up to thousands of cores.

"This enabled us to isolate the main issue and make the right design improvements," Ragavendra explained in an April 15 post. Because only one design iteration was needed, he noted that tooling, prototyping and testing costs were saved, as well as time – which can mean a lot. Engineers are expensive.

Speaking of time, the simulations themselves used to take days, but on Azure HPC servers – which are located in both Western North America and Southeast Asia – the boss engineer observed that it now takes hours. According to the blog, "large models with millions of degrees of freedom became routine and easily solved" with the switch to HPC resources.

Microsoft plans to build on the experience it has gained, adding more resources and enabling even greater scalability for multi-physics modeling.

“There is a huge opportunity to enable machine learning and AI in product creation,” Raghavendra wrote. ®

More about

TIP US OFF

Send us news


Other stories you might like