The move lands just as Intel was trying to convince the kernel community it still cared about upstream support.
At the end of November, Intel finally released the Habana Labs Gaudi 3 kernel driver code, years later than promised. That code missed the Linux 6.19 merge window thanks to delays, layoffs and a revolving door of driver maintainers, leaving it aimed vaguely at Linux 6.20 or even 7.0.
However, it turns out Intel has already walked away from maintaining the open-source user-space driver stack that the kernel code depends on.
Earlier this year, Chipzilla archived the SynapseAI Core repository on GitHub and slapped a blunt warning "README. PROJECT NOT UNDER ACTIVE MANAGEMENT. Intel will no longer maintain this project. Intel has ceased development and contributions, including, but not limited to, maintenance, bug fixes, new releases, or updates, to this project. Intel no longer accepts patches to this project.”
Anyone desperate enough to rely on it is told to fork it themselves and crack on alone.
SynapseAI Core provides the user-space APIs, the Synapse backend for executing code on Habana Gaudi hardware, thunk libraries and other essential plumbing needed to talk to the accelerator.
In short, it is the open-source user-space half that justifies the presence of an upstream kernel driver. Without it, the whole effort starts to look academic.
The code was originally open-sourced years ago, back when Habana was still independent.
At the time, upstreaming kernel drivers required a basic open-source user-space library to prove the hardware could actually be exercised.
With SynapseAI Core now effectively abandoned, that foundation has been kicked away.
It also puts the future of the Gaudi kernel driver in an uncomfortable position.
The issue has already been raised on the Linux Kernel Mailing List, along with the pointed observation that Gaudi 3 support never made it upstream before the user-space code was archived. That could yet become a blocker for getting Gaudi 3 into a future kernel cycle.
Given Chipzilla’s ongoing cost-cutting and engineering layoffs, the motivation is not hard to guess. Maintaining open-source code costs money, yet Intel still ships a closed-source Gaudi software stack.