Announcement

Collapse
No announcement yet.

NVIDIA R550 Linux Driver's Open Kernel Modules Performing Well On GeForce GPUs

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

  • avis
    replied
    Originally posted by tabicat View Post
    The fact that you haven't seen it means nothing.
    You definitely could have provided examples instead of being dismissive. It's so easy to insult the opponent for being unknowledgeable or stupid instead of providing actual data.

    Guess we are on the Linux forum where this type of communication is alright. It's not. It's horrible.

    Leave a comment:


  • tabicat
    replied
    Originally posted by avis View Post
    I've not seen any data in regard to the NVIDIA's own open source driver being used in any capacity.
    The fact that you haven't seen it means nothing.

    OpenRM is the only version of the driver that supports HMM, which Cuda uses. That is not an opinion. There are other GPL-only features that OpenRM supports, like DMABUF and Address Translation Services (ATS). Confidential Computing is another feature. The list will only grow as more kernel features are gated behind GPL symbols.

    Secondly, the way it exists (a ton of embedded binary data) it will never be merged with the kernel which further drives my point.
    No one ever said anything about trying to get OpenRM into the upstream kernel. OpenRM exists because it's the only way the driver can access GPL symbols.

    Leave a comment:


  • avis
    replied
    Originally posted by tabicat View Post

    This is definitely not true.

    The reason the open-source Linux driver (aka OpenRM) exists is because the driver needs access to GPL symbols in order to enable some advanced functionality, like HMM. The fact that Nouveau also uses the GSP binary blob is just frosting on the cake.
    I've not seen any data in regard to the NVIDIA's own open source driver being used in any capacity.

    Secondly, the way it exists (a ton of embedded binary data) it will never be merged with the kernel which further drives my point.

    Leave a comment:


  • tabicat
    replied
    Originally posted by Goddard View Post

    Why won't they include it in the kernel?
    Because the code as-is violates almost every Linux kernel coding standard and requirement. Any experienced kernel developer can take one look at the code and tell you that.

    Also, what isn't so well known is that the kernel side (aka CPU-RM) and the GSP side (aka GSP-RM) need to be compiled from the same header files. You can't mix-and-match kernel versions and GSP-RM versions. If even if CPU-RM were merged into the upstream kernel, it would only support a single version of GSP-RM. In fact, this is exactly the reason why Nouveau only supports GSP-RM version r535, and only one specific version of r535 at that.

    Leave a comment:


  • tabicat
    replied
    Originally posted by avis View Post
    AFAIK it only exists as a reference implementation to ease nouveau's implementation efforts.
    This is definitely not true.

    The reason the open-source Linux driver (aka OpenRM) exists is because the driver needs access to GPL symbols in order to enable some advanced functionality, like HMM. The fact that Nouveau also uses the GSP binary blob is just frosting on the cake.

    Leave a comment:


  • hancor
    replied
    Loaded the driver when it came out (Asus Strix RTX OC 4090, watercooled in custom loop), so far working nicely with the Asus PG42UQ monitor, and it seems happy with the latest Steam gaming updates. HDR 10bit feature is very nice & monitor colors are really rich.

    Leave a comment:


  • RealNC
    replied
    Originally posted by ptr1337 View Post
    I think it would be cool to directly integrate the kernel modules with a patch into the kernel
    If you mean upstream this into the kernel, then hell no. I want to be able to update my GPU driver without having to switch to non-LTS kernels.

    Leave a comment:


  • dragon321
    replied
    Originally posted by pong View Post
    Recently (this version or maybe the previous one) it was noted that to use some new CUDA features (unified heterogeneous memory) it was a requirement to be using the open kernel drivers. I don't know why or how that would be so, but it was mentioned in the new feature info release notes.

    It that was so, I suspect it could be possible that other disparate features could exist between the open kernel drivers installation and the other.
    This is actually a good point.​​​

    Originally posted by sarmad View Post

    Yes. Using it will force nVidia to keep it in good shape, and therefore will remain an up-to-date reference implementation for Nouveau developers, and an up-to-date kernel module for any future developer willing to build an open user space module to go with this open kernel module. Also, having an open kernel space module and a closed user space module is still better than having both modules being closed.


    Another good point but I guess they will move to it in the future anyway. Probably one of the reason why this driver exist is the fact that it will be able to use GPL only symbols in kernel and provide more functionality than proprietary driver can.

    Originally posted by pWe00Iri3e7Z9lHOX2Qx View Post

    One less 3rd party repo to add for some distros.
    It's useless without proprietary userspace but I guess you can get it in some different way. Since kernel module is open source I guess distributions can keep prebuilt version in repositories and avoid the need for recompilation during every kernel update.

    Anyway thank you all for answer, I'm going to try it on my PC. Since support for desktop GPUs is now considered as "certified" and performance is more or less the same I guess there is no good reason why not to.
    Last edited by dragon321; 07 March 2024, 02:30 PM.

    Leave a comment:


  • Rovano
    replied
    Originally posted by Anux View Post
    Why would you need to be convinced, didn't you read the article? You could try it yourself or read up on it and spare yourself much wasted time.

    You could use nouveau but that's not related to nvidia and also doesn't use the open kernel modules currently I think and it's slow.
    When you install an Nvidia binary blob, you get a choice of which module to use.
    I think the only practical difference for the average user is whether they have a card that already supports GPS or not.
    GSP guarantees card overclocking and other things.

    If your card supports GPS, it will be used regardless of whether you have the module open or closed.

    Nouveau(NVK) will be interesting from MESA 24.1. Vulkan 1.3 etc.​

    Leave a comment:


  • partcyborg
    replied
    Originally posted by Goddard View Post

    Why won't they include it in the kernel?
    It won't ever be accepted. It literally has inline compiled binary code as char arrays.

    NVIDIA Open GPU Kernel Modules Version Affects all versions Does this happen with the proprietary driver (of the same version) as well? I cannot test this Operating System and Version Affects all O...

    Leave a comment:

Working...
X