Five months after its initial release, TensorFlow may finally start finding use in large-scale artificial intelligence projects. The credit goes to the new iteration of the framework that Google rolled out yesterday, which introduces the ability to scale algorithms across hundreds and potentially even thousands of machines.

The functionality is made possible by the inclusion of a technology called gRPC that had likewise been created by the Alphabet Inc. subsidiary and became available under an open-source license last February. It’s a distributed workload orchestration engine built around the emerging HTTP/2 standard, yet another Google creation, that handles the logistics of spreading an application over a large number of disparate servers. But the software is not sufficient on its own to manage a large artificial intelligence cluster, which is why TensorFlow now also provides support for higher-level management frameworks like Kubernetes.

Rajat Monga, who leads development on the algorithm-building engine at Google, told WIRED that the reason the functionality wasn’t included in the original release is because his team sought to make the software accessible to the public as soon as it could. TensorFlow is based on the platform that the search giant uses internally to build its deep learning algorithms, he explained, which is built to operate in its highly specialized data center environment. As a result, it took a while for Monga’s team to adapt the framework for running on more conventional infrastructure.

The effort to make TensorFlow more scalable was probably also influenced by Microsoft Corp.’s release of a rivaling artificial intelligence framework called CNTK back in January. The system came with the ability to distribute algorithms across multiple machines out-of-the-fox, a feature that Redmond touted as a major advantage over the search giant’s alternative – and rightfully so. Now that the playing field has been leveled, the company will need to find other means of setting its offering apart from the pack.

That means artificial intelligence researchers can expect more advanced functionality to come their way, not only from Microsoft and Google but also all the other web giants that have joined the fray in recent months. China’s Baidu Inc. released a library for building voice recognition software in January, and Facebook Inc. followed suit the next month by open-sourcing several of its own internal algorithm development tools.

Image via Pixabay