Announcement

Collapse
No announcement yet.

Nouveau Patches Posted For Running On NVIDIA GSP-RM Firmware, Initial RTX 40 Ada Support

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • mdedetrich
    replied
    Originally posted by Amaranth View Post

    That's talking about how you build a `.run` package that contains binary code and a source code shim, just like the proprietary driver. They're selling it as a compile savings but it's probably just because they reused the design and installer from the proprietary version. The binary parts are compiled from source as a part of making the `.run`, it's not what I linked to in my last post.
    I know, but what you linked to in your last post is not evidence that they are putting a binary in https://github.com/NVIDIA/open-gpu-k...de_GA100.c#L29

    Leave a comment:


  • Amaranth
    replied
    Originally posted by mdedetrich View Post
    In fact according to the README, the only binaries they provide are the OS agnostic ones and this is out of convenience/practicality (i.e. they take ages to compile), see https://github.com/NVIDIA/open-gpu-k...kernel-modules
    That's talking about how you build a `.run` package that contains binary code and a source code shim, just like the proprietary driver. They're selling it as a compile savings but it's probably just because they reused the design and installer from the proprietary version. The binary parts are compiled from source as a part of making the `.run`, it's not what I linked to in my last post.

    Leave a comment:


  • mdedetrich
    replied
    Originally posted by Amaranth View Post

    Ok, fine, we can try again. Does this look like source code? Can you point to anywhere in the AMD driver that is included in the kernel that has something similar? Like I said, it looks like it's probably just firmware but that's still a program binary converted to byte array and dumped into the source code. That disqualifies it from being included in the kernel.

    To clarify in case you misunderstood the first time, I'm not saying that file is generated _from_ firmware, I'm saying that file most likely _is_ firmware.
    You mean this https://github.com/NVIDIA/open-gpu-k...ata.c#L84-L109 ? I don't believe it is what you think it is.

    In fact according to the README, the only binaries they provide are the OS agnostic ones and this is out of convenience/practicality (i.e. they take ages to compile), see https://github.com/NVIDIA/open-gpu-k...kernel-modules
    Last edited by mdedetrich; 29 September 2023, 03:27 AM.

    Leave a comment:


  • Amaranth
    replied
    Originally posted by mdedetrich View Post

    And you have issues with reading comprehension
    Ok, fine, we can try again. Does this look like source code? Can you point to anywhere in the AMD driver that is included in the kernel that has something similar? Like I said, it looks like it's probably just firmware but that's still a program binary converted to byte array and dumped into the source code. That disqualifies it from being included in the kernel.

    To clarify in case you misunderstood the first time, I'm not saying that file is generated _from_ firmware, I'm saying that file most likely _is_ firmware.

    Leave a comment:


  • mdedetrich
    replied
    Originally posted by Amaranth View Post

    Ok, so you're willfully obtuse then. Understood.
    And you have issues with reading comprehension

    Leave a comment:


  • Amaranth
    replied
    Originally posted by mdedetrich View Post

    Are you talking about NVidia's open source driver, because we were in both cases talking about source code not binaries (and in both cases the source code is pregenerated from some firmware).

    Obviously NVidia's blob is closed source, but thats a separate discussion/point.
    Ok, so you're willfully obtuse then. Understood.

    Leave a comment:


  • mdedetrich
    replied
    Originally posted by Amaranth View Post

    AMD's driver has generated header files for registers and such, Nvidia's driver has compiled binaries without any public source code embedded in it. It looks like those might be firmware but that's still not okay for upstream, it has to load those from disk.
    Are you talking about NVidia's open source driver, because we were in both cases talking about source code not binaries (and in both cases the source code is pregenerated from some firmware).

    Obviously NVidia's blob is closed source, but thats a separate discussion/point.

    Leave a comment:


  • toughy
    replied
    Originally posted by zexelon View Post
    This is awesome news!

    TBH I don't see a future where Nouveau holds any sort of performance edge to Nvidia's in house, but hey, a better default experience is still awesome!
    Oh, after a decade of reclocking firmware conspicuously missing from NVIDIA, at this point I would be so happy with just 85-90% of the performance of the proprietary driver ! No performance edge needed

    Leave a comment:


  • toughy
    replied
    Can someone post build instructions, maybe on a blog ? I just can't wait for the full kernel release...

    Leave a comment:


  • Amaranth
    replied
    Originally posted by mdedetrich View Post

    There is nothing wrong with pre-generated/compiled code, software projects do it all the time. Whether to include that precompiled code in the tree/vcs or integrate the generation into the build system is an open question and there isn't a right/wrong answer either way (and yes this also includes if the generation of the code is also open).

    Also FYI, AMD driver also has a lot of pre-compiled code in the form of headers which comes from proprietary sources. Not sure how comparable it is, but this isn't that concerning/shocking as you seem to imply.
    AMD's driver has generated header files for registers and such, Nvidia's driver has compiled binaries without any public source code embedded in it. It looks like those might be firmware but that's still not okay for upstream, it has to load those from disk.

    Leave a comment:

Working...
X