Daniel Goldin has an impressive resume. The 75-year-old spent over a quarter century in the aerospace industry during the first leg of his career, went on to become the director of NASA and is now returning to the fold as the head of a newly launched startup. KnuEdge Inc., as the outfit is called, hit the scene today with a homegrown processor specifically designed to run machine learning algorithms.

Dubbed KNUPATH, the chip sports 256 cores and 16 bidirectional I/O paths that provide 320 Gbs of throughput. It’s also well-equipped to run in large environments, with the startup claiming that a single deployment can scale to over half a million nodes while keeping inter-rack latency at around 400 milliseconds. The end-result is an up to sixfold improvement in the performance of artificial intelligence workloads per watt consumed.

That’s not quite as impressive as the ASIC that Alphabet Inc. built to accelerate its machine learning projects, which is described as an entire order of magnitude faster than commercial alternatives. But KNUPATH still handily beats anything that organizations can currently buy on the open market. KnuEdge developed the chip after finding the performance of existing CPUs and GPUs inadequate to support its main offering, a speech recognition engine called KnuVerse that took nearly a decade to develop.

The software sets itself apart by providing the ability to parse voice input even when there is a large amount of background noise that alternative tools can’t handle. KnuEdge says that the technology could help consumer-facing businesses make support calls less of a hassle in situations where a TV is playing in the background or the caller is outdoors. Moreover, KnuVerse also the potential to greatly improve voice authentication: The startup claims it’s able to verify a person’s identity after just a few words.

KnuEdge plans to make the technology available from the cloud as a freemium service and sell its KNUPATH processors to organizations that place a premium on performance. The startup says that it’s already generated about $20 million in revenue so far, which comes on top of the $100 million it raised from investors prior to exiting stealth. Given the fast-growing enterprise interest in artificial intelligence technology, its war chest can be expected to swell even further over the coming months and quarters.

Image via Geralt