Intel's Habana Gaudi 2 Accelerator Linux Driver In Good Shape

Since June of last year shortly after Intel announced the Gaudi 2 AI accelerator they began posting the open-source driver patches for Gaudi 2 with the mainline Habana Labs driver. That support landed in Linux 6.0 and since then they've continued refining that support for this new processor support for deep learning training and inference workloads.

With succeeding kernel updates Intel has continued improving the Gaudi 2 support in the Habana Labs kernel driver, but it hasn't been clear how stable or complete the support has been due to lack of hardware access and no public communications yet over what pieces may be missing or how stable the support is with this continually-improving open-source code.

Today is the first confirmation that the Intel Gaudi 2 support is pretty much in good shape right now if using the latest mainline Linux kernel driver. Longtime Habana Labs Linux driver developer/maintainer Oded Gabbay commented in today's pull request:

"As Gaudi2 is pretty much stable, this contains mostly bug fixes and small optimizations and improvements."

So with the current Linux 6.4 kernel the Gaudi 2 support should be "pretty much stable" while for the upcoming Linux 6.5 cycle is just bug fixes and smaller optimizations.

That comment was via this pull request of the Habana Labs accelerator driver improvements that were submitted today for DRM-Next to in turn make their way into the mainline kernel once the Linux 6.5 merge window opens around the end of the month.

In addition to the Gaudi 2 support within the Linux kernel, over in user-space their open-source SynapseAI Core has Gaudi 2 support already in place too. Intel's Habana Labs AI accelerators continue to be the most open-source, upstream accelerator option currently available for AI training and inference.