An NVIDIA representative addresses inquiries on transitioning drivers to open kernel modules

Andy Ritger (Andy Ritger), who heads the development of drivers for Unix-like OS at NVIDIA, answered to questions raised regarding the transition of NVIDIA's proprietary drivers to use open Linux kernel modules by default on systems with GPU-based microarchitectures starting with Turing.

  • When asked about achieving parity in the functionality of open and proprietary drivers, an NVIDIA representative replied that in driver release 560, open source kernel modules will be approximately equal in functionality to proprietary modules. Among the limitations, the impossibility of using a dynamic energy management mechanism is noted RTD3 (Run Time D3) with open kernel modules on older generations of GPUs preceding Ampere (proprietary modules additionally support GPUs on the Turing microarchitecture).

    The capabilities of open and proprietary modules related to GPU initialization and power management will be quite similar in the NVIDIA 560 drivers, and over time, work will continue to achieve full parity in this direction. Release 560 will also address some long-standing issues in open modules, such as those related to the use of VRR (Variable Refresh Rate) on laptops.

  • NVIDIA has no plans to push for open modules to be included in the main Linux kernel.
  • When asked about support for open Nouveau and NVK drivers, an NVIDIA representative replied that for production use the company recommends using proprietary drivers and separately supplied open kernel modules. The company has previously made attempts to provide developers with nouveau and nvk documentation on NVIDIA chips and hardware interfaces and provided individual patches, but the assistance provided is modest enough to be called support.
  • When asked about the possibility of support in the open Nouveau driver for proprietary NVIDIA components running in user space, such as CUDA, AI, RT/PT, DLSS and Optix, it was indicated that this is not yet possible and these components cannot work when using the nouveau kernel module . Whether this will be possible in the future is not yet clear.
  • It is noted that with nouveau it is also not yet possible to use features related to configuration and monitoring (nvapi/nvidia-smi). At the same time, a previous license change for nvapi API definitions allowed the wine and proton projects to prepare their own implementations of some nvapi elements used in games.
  • When asked about the participation of NVIDIA employees in the development of nouveau, it is indicated that such participation is still insignificant, but some employees are already contributing to the development of Nouveau and participating in discussions.
  • NVIDIA is not going to provide open kernel modules for older GPUs before the Turing generation. Users of Volta and older GPUs should continue to use the proprietary modules.
  • When asked about the company's plans to open up components of the driver stack for GPUs running in user space, an NVIDIA employee who answered the question said he had not heard of such plans.
  • The decision to use public modules by default is driven by the desire to simplify testing and reduce the overhead of having to retest public and private modules.

Thanks for reading: