There are signs that Intel has encountered new setbacks in the open source software ecosystem of its Gaudi accelerator: SynapseAI Core, the open source project responsible for Gaudi user space support, has been archived and ceased maintenance, casting a shadow over the future of related kernel drivers.

Intel just released the Habana Labs kernel driver code for the Gaudi 3 accelerator at the end of November and is preparing to submit it upstream for integration into the Linux kernel mainline. Previously, this open source work was repeatedly hindered by layoffs and personnel turnover within the company, and the driver maintainer was changed many times, resulting in long-term delays in kernel support for Gaudi 3. Due to the late code release, this driver failed to catch up with the Linux 6.19 merge window and could only target Linux 6.20 or even 7.0 versions.
The bigger problem, though, is the user-space software stack. Intel archived SynapseAI Core on GitHub earlier this year, and clearly marked the project description in the project description that the project is "no longer actively maintained by Intel." The company will no longer provide support for development, repairs, updates, or accept patches. It also recommends that users with ongoing needs fork the code themselves and maintain it independently. SynapseAI Core is a key library that provides user-space APIs around Gaudi hardware. It includes the Synapse backend, user-space thunk library and other components required to execute code. It is the basic software layer that utilizes accelerators and upstream kernel drivers.
SynapseAI Core was originally open sourced when Habana Labs was still an independent company. At that time, in order to meet the upstream requirements of the Linux kernel, it opened up the basic user mode library used to cooperate with the operation of the kernel driver in order to complete the testing and functional verification required for the driver to enter the mainline. Now that this core user space project is archived and no longer maintained, it means that the upstream prospects of the Gaudi accelerator driver on the kernel side are substantially threatened in the absence of supporting open source user space support.
This situation was also brought up for discussion on the Linux Kernel Mailing List (LKML), with comments stating that Gaudi 3 support was never actually completed and merged upstream until SynapseAI Core was archived. If the current situation does not change, then in the absence of an active open source user space stack, Gaudi 3 support will likely face obstacles in successfully entering the mainline in the next kernel development cycle.
From the outside world, this change is closely related to Intel's recent cost control and the layoff of the engineering team. The company is obviously unwilling to continue to invest resources in maintaining this part of the open source code. In addition, there is still a closed-source Gaudi software stack within Intel, and Gaudi 3 is regarded as one of the final nodes of this product line. Stopping the open source maintenance of SynapseAI Core is likely to be part of the overall "finishing" and gradual fading out of related software support. It remains to be seen how the Gaudi open source ecosystem will develop in the future.