Announcement

Collapse
No announcement yet.

NVIDIA Still Working On Linux 3.11+ Support

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

  • NVIDIA Still Working On Linux 3.11+ Support

    Phoronix: NVIDIA Still Working On Linux 3.11+ Support

    In mid-October I had written how AMD's Catalyst driver surprisingly beat NVIDIA to modern Linux support. While NVIDIA is usually first to support new kernel releases, AMD won in shipping "out of the box" Linux 3.11 and 3.12 compatibility. NVIDIA, however, has devised a workaround and will be coming up with a more proper long-term solution...

    http://www.phoronix.com/vr.php?view=MTUwMTk

  • #2
    Middle finger by Linus still holds true.

    Nvidia need this function, and they had all the time in the world to voice their needs... Where where they?

    Comment


    • #3
      Well, at least we now have a stopgap measure that is more or less safe to use. Just in time for openSUSE 13.1, too.

      Comment


      • #4
        What are people doing on Ubuntu 13.10?

        Comment


        • #5
          Am I missing something or should the article say 3.12 ?

          nvidia blob working fine for me on Arch Linux with 3.11

          Code:
          me@upstairs ~ $ uname -a
          Linux upstairs 3.11.6-1-ARCH #1 SMP PREEMPT Fri Oct 18 23:22:36 CEST 2013 x86_64 GNU/Linux
          
          me@upstairs ~ $ pacman -Qi nvidia
          Name           : nvidia
          Version        : 325.15-10
          Description    : NVIDIA drivers for linux
          Architecture   : x86_64
          URL            : http://www.nvidia.com/
          Licences       : custom
          Groups         : None
          Provides       : None
          Depends On     : linux>=3.11  linux<3.12  nvidia-libgl  nvidia-utils=325.15
          Optional Deps  : None
          Required By    : None
          Optional For   : None
          Conflicts With : nvidia-96xx  nvidia-173xx
          Replaces       : None
          Installed Size : 3963.00 KiB
          Packager       : Tobias Powalowski <tpowa@archlinux.org>
          Build Date     : Mon 14 Oct 2013 07:39:41 BST
          Install Date   : Tue 15 Oct 2013 09:29:00 BST
          Install Reason : Explicitly installed
          Install Script : Yes
          Validated By   : Signature
          
          me@upstairs ~ $ glxinfo |grep "vendor string"
          server glx vendor string: NVIDIA Corporation
          client glx vendor string: NVIDIA Corporation
          OpenGL vendor string: NVIDIA Corporation

          Comment


          • #6
            Using this patch I presume:
            Code:
            --- a/kernel/nv-linux.h
            +++ b/kernel/nv-linux.h
            @@ -957,7 +957,7 @@ static inline int nv_execute_on_all_cpus(void (*func)(void *info), void *info)
             #endif
             
             #if !defined(NV_VMWARE)
            -#define NV_NUM_PHYSPAGES                num_physpages
            +#define NV_NUM_PHYSPAGES                totalram_pages
             #define NV_GET_CURRENT_PROCESS()        current->tgid
             #define NV_IN_ATOMIC()                  in_atomic()
             #define NV_LOCAL_BH_DISABLE()           local_bh_disable()
            with fuzz +1 line

            Comment


            • #7
              Meanwhile Catalyst beta still builds fine on the latest 3.12-rc7 kernel.

              So glad I went back to AMD.

              Comment


              • #8
                Lots of posts so far criticizing Nvidia for this, but I can't imagine too many people are affected by this. Anyone with 128+ GB is most likely on an LTS kernel, and with 3.2 EOL'ing in 2016, they have plenty of time. (Interestingly, 3.4 EOLs next year and 3.10 in 2015 - does anyone know how LTS periods are decided?).

                Comment


                • #9
                  Dang, my laptop with 129GB of RAM is screwed...

                  On a more serious note, at least we now know what the hold up was. Most major distros already had their own stopgap patches in place for new/upcoming releases, so this is a non-issue for most people. Still, it will be nice to see a proper solution somewhere down the line, not that the average user is going to reach over 128GB of RAM any time soon, or even a rich user for that matter.

                  Comment


                  • #10
                    Originally posted by sirdilznik View Post
                    Dang, my laptop with 129GB of RAM is screwed...

                    On a more serious note, at least we now know what the hold up was. Most major distros already had their own stopgap patches in place for new/upcoming releases, so this is a non-issue for most people. Still, it will be nice to see a proper solution somewhere down the line, not that the average user is going to reach over 128GB of RAM any time soon, or even a rich user for that matter.
                    Yeah, doubt any home computer user has that much RAM. Most I seen for a personal home computer was 16GB. Even if they were running a home server, still wouldn't need 100+GB.

                    I've only seen datacenters run that amount of RAM since they need it.

                    Regardless, supporting latest is nice. 3.12 is so much better than most previous versions I've used.

                    Comment


                    • #11
                      Originally posted by sirdilznik View Post
                      Most major distros already had their own stopgap patches in place for new/upcoming releases, so this is a non-issue for most people.
                      Their patches were stabs in the dark, reporting lies to the driver. This latest official patch at least makes sure that everyone below 128 GiB are not affected; the unofficial ones don't.

                      Comment


                      • #12
                        Originally posted by synaptix View Post
                        Yeah, doubt any home computer user has that much RAM. Most I seen for a personal home computer was 16GB. Even if they were running a home server, still wouldn't need 100+GB.

                        I've only seen datacenters run that amount of RAM since they need it.

                        Regardless, supporting latest is nice. 3.12 is so much better than most previous versions I've used.
                        Haswell already supports 32GB memory and some Clevo resellers are offering 32GB memory as a customizable option (at a very huge price tag though).

                        Soon it will be 64GB memory...then 128GB...then 256GB...and then Nvidia will have a problem again.

                        Comment


                        • #13
                          Originally posted by Sonadow View Post
                          Soon it will be 64GB memory...then 128GB...then 256GB...and then Nvidia will have a problem again.
                          I don't think that you can buy a system with 256 GB RAM and not break the bank any time soon. Most consumer mobos max out at 32 GB currently (4*8GB). Even entry level server mobos that support registered memory like the ASUS KCMA-D8 would need to be equipped with 32 GB RDIMMs which are quite expensive still ($500+ a pop).

                          By the time 256 GB RAM systems become affordable, I am sure NVidia will have found a way to make the driver work properly with newer kernels.
                          Last edited by chithanh; 11-03-2013, 12:06 PM. Reason: I fail at math

                          Comment


                          • #14
                            Bleh Nvidia and Amd just need to open the source code of their binary blobs and get the code into the kernel and this would not be an problem.


                            But no they think they have some uber national secrets embedded in their code LOL

                            Comment


                            • #15
                              Originally posted by Sonadow View Post
                              Haswell already supports 32GB memory and some Clevo resellers are offering 32GB memory as a customizable option (at a very huge price tag though).

                              Soon it will be 64GB memory...then 128GB...then 256GB...and then Nvidia will have a problem again.
                              Sonadow, You are being silly / not thinking about this clearly / are constructing a very poor argument. Please use some common sense;

                              1st. The current patch is a *temperary workaround*, until Nvidia resolves the issue. (which i imagine wil be solved in a release or two - ie: a short time from now).

                              2nd. By the time 128+ Ram becames "the norm" for even a minority of average users, this issue will be long gone and probably long forgotten, as well.... being as Nvidia is already working on it and i can't imagine this is going to take YEARS to solve ~ which is the time-frame that would be required, in order for your argument / point to make *any sense*, at all! (which it doesn't).

                              3rd. Intel supporting 32GB or 64GB is nothing new, nor is this likely a new concept for Nvidia, either. There are 2nd Gen i7s (Extreme editions) that can support 64GB of RAM, with the proper MOBO, etc...and obviously Intel XEONs have support large amounts of RAM for years. (ie: 128GB+ of RAM). Not to mention that it isn't uncommon to See Mac Pros that exceed 32GB of RAM, in professional environments (Like Studios). Example; Article / reference; http://macperformanceguide.com/blog/...GB-in-use.html <note the 2009/10 models support well above 32GB of RAM, regardless of H/W configuration, even in MacOSX Mountain lion>

                              anyway, i don't see this as a problem at all. Something changes in the kernel that Nvidia was using. now they will adapt their driver - no big deal.

                              Comment

                              Working...
                              X