No announcement yet.

NVIDIA Pushes 62MB Of GSP Binary Firmware Blobs Into Linux-Firmware.Git

  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    It really is a big joke from nvidias side. Hopefully that bullshit will be packaged into a separate package on most distros...


    • #32
      That smells like some lazy development. I can bet they can split that into 1 common code GSP binary and then smaller GPU-specific GSP binaries.


      • #33
        looks like amd did it way better with atombios approach. but nvidia had to reinvent the wheel and failed.


        • #34
          I don't get it. Neither for Nvidia nor AMD. Why for the name of Stan this firmware's do not present on GPU nvram from the start? Why do the they need to be loaded at runtime?

          Then they can be a separate package or even just delivered by fwupdate tool.


          • #35
            62MB is pretty big but it's not that big issue I guess. Much bigger issue is the fact that GSP ABI is not stable. So any firmware update will require changes in Nouveau driver. And if in future NVIDIA will drop support for old GPU in GSP then Nouveau will need to keep older GSP to support older cards which will take even more space.


            • #36
              Originally posted by stormcrow View Post

              Which is why it's never been a good idea to use the EFI partition for anything more than a boot stub (some Linux distros being the only OSes stupid enough to do so that I'm aware of). The distributions that use it for the entire boot chain are just asking for trouble - and will eventually find it for various reasons: some technical like the lack of any kind of internal data integrity in FAT nor permissions and security features - you can't count on motherboards to support more than the base requirement FS support (meaning no more than FAT), some because they didn't think the decision through in just how big the boot chain may grow in the future. It's not feasible for users to have to try to figure out how to resize or move their EFI partition because a year or two after install the boot chain grew way more than anticipated.
              The UEFI standards allow use of any filesystem for the ESP, you just* need to have the necessary EFI drivers available. In some cases, they don't exist. However, the UEFI standards mandate that the UEFI version of FAT (which is subtly different to standard FAT) is supported, so you always have that as a fallback.

              For most practical purposes, people use UEFI FAT.

              Pete Batard has done an excellent job in providing Free Software read-only EFI drivers, derived from GNU GRUB** ones.

              Free Software EFI/UEFI file system drivers

              Github: EfiFs - EFI File System Drivers

              *The practical problem is getting the necessary drivers incorporated in your PC's firmware.

              **If you use GRUB as a boot manager/bootloader you can load filesystem drivers as GRUB modules, and thereby, during the boot process, before you have loaded linux, access extra resources on disks or partitions other than the ESP, which is what I did for years. It is very easy to do manually, but not so easy to incorporate into the standard update-grub workflow. There are features and bugs that trip up the non-expert. It means that there are mechanisms by which the amount of 'stuff' forced to be resident on the UEFI FAT ESP can be minimised.


              • #37
                Originally posted by avis View Post

                That deserves yet another NVIDIA haters are gonna hate. You're free to enjoy discrete AMD and Intel GPUs and spare us from your insightful comments. No one fucking cares what companies you like or not.
                OK Nvidia fanboy!
                How about you spare us from defending Nvidia from rightful critique like you have bout shares from them?
                And just so you know, I'm already enjoying AMD and Intel GPUs as I've ditched Nvidia many years ago for the same attitude that they still have today.


                • #38
                  Originally posted by Danny3 View Post

                  OK Nvidia fanboy!
                  How about you spare us from defending Nvidia from rightful critique like you have bout shares from them?
                  And just so you know, I'm already enjoying AMD and Intel GPUs as I've ditched Nvidia many years ago for the same attitude that they still have today.
                  If you're already using these super companies GPUs, why are you spamming this thread? Your input is completely wasteful.

                  Again, you cited me saying NVIDIA haters gonna hate and proceed to do just that. The irony is completely lost on you. What a shame.

                  And lastly, I have never defended NVIDIA. You're welcome to prove otherwise. Being rational about them is not equal to defending them. Not hating them is not defending them. I know it's difficult to be a rabid fan, but other people don't necessarily divide the world into things you hate or love. I just exist. I don't have strong emotions about companies whose only concern is profits. If you believe AMD and Intel are somehow different, you need to reevaluate your worldview.

                  And if you believe AMD and Intel are somehow more user friendly to open source than NVIDIA, you also need to reevaluate your worldview. Both companies support Linux just enough to make it work. They invest up to 1000 times more money and resources into supporting Windows. Their Windows drivers dwarf their Linux drivers in terms of features. Under Windows HWiNFO64 reports basically nearly 50 times more data about Intel/AMD CPUs/GPUs than what's available under Linux. They have a ton more features than what Linux drivers provide. Open Source fans always forget about that for some reasons. It's pathetic and despicable at the same time. At the same time NVIDIA does not pretend they support Linux any more than they support Windows but they are the only company who shares their codebase between Linux and Windows, which means their driver quality will forever be a lot higher than what Intel or AMD could ever achieve with the kernel drivers and MESA. Why? Because NVIDIA cares and Windows has WHQL. Linux has nada. A broken Intel/AMD GPU commit in the kernel? Who cares? A few releases later you'll get your Intel/AMD GPU working again. No QA/QC whatsoever (TBO Intel has some sort of automated testing but it still results in regressions here and there).

                  Enjoy your Linux supremacy except there's none. And don't BS me with "AMD/Intel GPU drivers are open, anyone can fix them". That's never been the case. Valve helps AMD only because of the Steam Deck but Valve is not "anyone". GPU drivers nowadays are a lot more complicated than the kernel itself. Process/memory/interrupts/PCI-E bandwidth scheduling + their own compiler and shading language/etc/etc/etc. The average programmer will need months if not years of extensive deep dive just to start hacking them.

                  Lastly, I've never called you or anyone here an AMD/Intel fanboy. Maybe you could be more respectful to other people. Being anonymous doesn't mean it's OK to be an asshole.
                  Last edited by avis; 09 November 2023, 07:46 AM.


                  • #39
                    Originally posted by qarium View Post

                    yes this is really insanity. why do all the people think this is a good idea ?
                    It's not a problem for people using custom kernels and AMD gpus. Noobuntu users have to pay for the extra bandwidth.


                    • #40
                      Can someone tell me where is the problem with the size?
                      The article mentions that it is lower than the AMD counterpart so i cant wrap my head around some of the fanboy/hater comments.
                      Also why would it need to be included in initramfs?
                      Isn't the filesystem already accessible at a time GPU drivers are loaded?
                      It should be as i was running without initramfs until deciding to use raid for EFI and boot partitions.

                      Note that i will not buy nVidia GPU because they dictate when, where and how the GPU can be used via their drivers.
                      See the VM blocking, blockchain compute limiting and EULA disallowing running in datacenter for gaming GPUs.
                      Don't want to be a victim of artificial product segmentation.