The good news is that Intel tonight posted a pull request for open-source Gaudi 3 accelerator support for the mainline Linux kernel! The bad news is that it’s coming quite late in the product cycle, much later than the former excellent Habana Labs open-source track record, and their hopes of squeezing this code into the Linux 6.19 kernel may be dashed.
Going back to even the pre-Intel acquisition days, the Habana Labs accelerator driver had a good open-source track record. Initially it started off a bit rough until they had open-sourced their user-space bits and ultimately their compiler and other user-space software. But after that they were excellent open-source stewards friendly with the mainline kernel and punctually enabling new hardware support, helping craft the Linux kernel accelerator “accel” subsystem under the Direct Rendering Manager area, etc.
But things began to fall apart after Intel went through several rounds of Habana Labs driver maintainers due to various engineers leaving the company. They hadn’t upstreamed any Gaudi 3 accelerator support and their other contributions dried up to this driver. Last year I was told they’d post their Linux driver support in October (2024) and that never panned out. Only tonight did they post their patch series enabling Gaudi 3 support some 13 months after that original timeline and after Gaudi 3 has already been shipping for more than one year.
So tonight there is a pull request pending for Gaudi 3 support for the Habana Labs upstream Linux kernel driver plus a variety of other improvements. But this pull request to DRM-Next was posted an hour after the Linux 6.18 kernel release. Typically the cut-off for new feature material to DRM-Next is around the -rc6 timeframe, or roughly two weeks ago for new feature material desiring to be into Linux 6.19. So that alone makes it highly questionable whether the DRM maintainers will accept a pull request this pronounced and late. Even if the DRM/accel maintainers accept it, Linus Torvalds may have a fit seeing new patches queued what will be just days old.
Making matters worse is this is a massive pull request. The pull request acknowledges there is a lot of code from the past few years they are only now upstreaming:
“We are excited to continue the upstreaming of the accel/habanalabs driver. While maintaining the momentum with Gaudi2 fixes, we now introduce Gaudi3 support, covering the development work done during the 2021-2024 period.
This tag contains habanalabs driver changes for v6.19, continuing the previous upstream work from tags/drm-habanalabs-next-2025-05-25. It includes improvements in debug and visibility, general code cleanups, and the finalization of Gaudi3 support, closing remaining gaps relative to Gaudi2. This is backed by foundational features such as the MMU v3 architecture, ODP (On-Demand Paging), and enhanced security infrastructure.”
In covering patches collected over the past ~4 years, there is a lot of them and the pull request includes 295,357 lines of new code while removing/modifying just 1,192 lines of code. A lot is header files but still for submitting something so big with the Linux 6.19 merge window now technically open makes it highly unlikely that any reasonable code review or oversight could be performed in time for the two-week Linux 6.19 merge window.
So this pull request is great in finally seeing the Intel Gaudi 3 accelerator open-source driver pull request for the mainline kernel, but it’s very debatable if it will actually be honored for the Linux 6.19 merge window. Even if it is accepted for Linux 6.19, that stable kernel won’t be officially out until February where as Intel “launched” Gaudi 3 back in September 2024.
