Comment 16 for bug 1897910

This bug was fixed in the package nvidia-graphics-drivers-450-server - 450.80.02-0ubuntu0.18.04.3

---------------
nvidia-graphics-drivers-450-server (450.80.02-0ubuntu0.18.04.3) bionic; urgency=medium

  * debian/templates/dkms_nvidia.conf.in:
    - Drop buildfix_kernel_5.8.patch
  * debian/templates/dkms_nvidia.conf.in,
    debian/templates/nvidia-dkms-flavour.install.in,
    debian/templates/nvidia-dkms-flavour.postinst.in,
    debian/templates/nvidia-dkms-flavour.prerm.in,
    debian/templates/nvidia-kernel-source-flavour.install.in:
    - Change DKMS directory and name adding "-srv". This should fix
      kernel builds now that the 450 and the 450-server drivers share
      the same version.

nvidia-graphics-drivers-450-server (450.80.02-0ubuntu0.18.04.2) bionic; urgency=medium

  * debian/templates/control.in:
    - Add missing comma in cuda provides (LP: #1897910).

nvidia-graphics-drivers-450-server (450.80.02-0ubuntu0.18.04.1) bionic; urgency=medium

  * New upstream release (LP: #1897910):
    - Improvements:
      o Added a new system-level interface using nvidia-capabilities
        via /dev for Multi-Instance GPU (MIG) on NVIDIA A100.
        For more information on these new device node capabilities,
        refer to the MIG user guide.
     o Added support for CUDA compatibility for CUDA 11.1, where CUDA
       11.1 applications will work on supported GPUs when used with
       drivers >= 450.80.00 on Linux. Refer to CUDA Compatibility for
       more information.
    - Fixed Issues:
      o Various security issues were addressed. For additional details
        on the med-high severity issues, review the NVIDIA Security
        Bulletin 5075.
      o Fixed an issue where, when NVLink training encounters a fatal
        error, link training is still considered successful resulting
        in broken nodes assigned to clients. An escape case in the
        NVLink link state reporting caused the driver to report the
        links as "Active" and functional even though the link had
        received a fatal error. This fatal error does not allow
        traffic to be sent over the link, and thus would cause P2P
        apps to fail. This is a reporting issue, but not a
        functionality issue. The lacking reporting case was
        addressed, so the failing link state is properly reported.
      o Fixed a software race condition that caused link training
        failure (CUDA_ERROR_LAUNCH_FAILED). A race condition within
        the driver polling code caused a false failure in some of
        the link training steps. This causes the driver to not
        continue link training even though training actually was
        successful up to that point. The fix corrects these
        sequences so that these false failures no longer happen.
  * debian/templates/control:
    - Provide cuda 11.0-1 and 11.0-1.

 -- Alberto Milone <email address hidden> Fri, 02 Oct 2020 18:58:37 +0200