Intel wants you to know about oneAPI – an open programming model for accelerated computing that lets developers write code without needing to tailor their application based on which company made the hardware that code will run on.

First announced in late 2019 as a project within Intel, oneAPI first emerged as a specification. Intel completed the project's first implementations, all backed by open-source repositories and open specifications, and "then we started building it out," Intel's Joe Curley, VP and GM of software products and ecosystem unit, told SDxCentral.

Early on, the project's technical advisory board was made up of representatives from Intel and its competitors. oneAPI received significant technical contributions from Xilinx, Curley said, in terms of how to build out a truly vendor-neutral specification. The project is now owned by its community, and while Intel could introduce "even greater written or formal structures around that, it is already an open, community-driven specification," he noted.

To that point, oneAPI is "a pretty simple idea that helps [enterprises] take accelerated computing and make it productive globally, regardless of the vendor." The programming model's "abstract, device-independent programming language" is called SYCL, which Curley described as modern C++ with accelerator languages and standardized library interfaces built in. That means developers can use one consistent language to program any vendor's hardware more effectively.

"Of course, we've built the tools out for our internal accelerators. So you can write code and you can have it work on a new fourth-generation Xeon or on one of our GPUs," he said. But that process is just as easy for other accelerators. Fujitsu, which uses an Arm-based processor, relied on oneAPI shortly after the project's initial launch in 2019 to help a customer run artificial intelligence (AI) on a supercomputer, breaking the MLPerf record at the time.

Curley credits that success to the standardized interface of oneAPI's neural network library, which "allows the Arm libraries to be found underneath it. Relatively quickly, ... they were able to – with very little help" from Intel or Arm – "get that up and running and deployed to their users in a matter of weeks," he said.

Besides time-to-value, organizations using this programming model see "fundamental benefits" like cost savings and the freedom to use whatever hardware accelerator they want without "getting locked into any one brand because of the programming model," he argued.

Although Intel is a company that could benefit from locking customers in, its work with oneAPI "creates a larger serviceable market," both for Intel and other vendors. "We want to ... defragment the market ... so the people can choose the right hardware for themselves rather than having the software they write always choose it for them," Curley said.

A defragmented market leaves Intel's success up to the quality of the hardware it builds, not its ability to trap customers. And "it allows us to bring our innovations to market more quickly, with less friction and hopefully faster adoption," he added.