View Single Post
Old 10-21-05, 08:01 AM   #1
NVIDIA Corporation
zander's Avatar
Join Date: Aug 2002
Posts: 3,740
Exclamation If you have a stability problem, PLEASE read this first

Please check the following before reporting stability problems:
  • If you are not using the most recent stable graphics driver release with support for your GPU(s), please update your driver installation. If you are using the latest driver release and are experiencing a stability problem you believe (or suspect) may be a regression, please try an older stable graphics driver release.

  • Many deployed Linux 2.4 and older Linux 2.6 kernels have known bugs in their implementations of kernel interfaces relied upon by the NVIDIA Linux graphics driver, such as change_page_attr(). If you are experiencing stability problems, please be sure to try a recent stable Linux kernel (vanilla or supplied by your distributor).

  • Verify that your system has an up-to-date SBIOS installed. Mainboard and/or OEM vendor supplied SBIOS update frequently include important bug fixes, many of which greatly improve system reliability, especially on new and/or complex system configurations.

  • If your graphics card vendor provides video BIOS updates, make sure you have the most recent release installed.

  • If you are seeing severe stability problems and you are using a Linux 2.6 SMP kernel on a system with multiple processors (or processor cores) in combination with more than one GPU, please search the output of `dmesg` for the presence of the message below after the system has just been started:

    If this message is present, please boot the system with the pci=nommconf kernel parameter and check if the stability problems continue to reproduce.

  • If your system is equipped with a dual-core processor, booting with the idle=poll and/or maxcpus=1 kernel parameters may improve reliability with some Linux kernels.

  • Similarly, forcing the clock source to HPET by booting with the clocksource=hpet kernel parameter has been reported to help on some systems. In order for this to work, you may need to enable HPET in your SBIOS and kernel configuration.

  • If you are using an AGP graphics card, please test setting the NvAGP configuration option to 0 in your X configuration file. If this eliminates the stability problems, you are most likely experiencing a problem outside of the NVIDIA graphics driver stack (e.g. in the SBIOS, Linux kernel, the AGPGART driver, the chipset or possibly other system hardware components).

  • If you are using a Linux/x86-64 2.6 kernel and see the warning message below during the installation and/or in the system log file(s) when the NVIDIA kernel module is loaded, please upgrade your kernel to Linux 2.6.11 or a more recent stable Linux 2.6 release. Linux/x86-64 2.6 kernels < Linux 2.6.11 have an accounting bug in their implementation of the change_page_attr() interface that can trigger a kernel BUG().

    NVRM: Your Linux kernel has known problems in its implementation of
    NVRM: the change_page_attr() kernel interface.
    NVRM: The NVIDIA graphics driver will attempt to work around these
    NVRM: problems, but system stability may be adversely affected.
    NVRM: It is recommended that you update to Linux 2.6.11 (or a newer
    NVRM: Linux kernel release).

    Current NVIDIA Linux/x86-64 graphics driver releases work around this problem by disabling use of the change_page_attr() interface on these kernels.

  • If you see warning messages similar to those below in the system log file(s) when starting the X server or OpenGL applications, please update the Linux kernel installed on the affected system. If the problem persists with the latest distribution kernel, please contact your distributor and submit a bug report to Please also read the section on Cache Aliasing in the NVIDIA Linux graphics driver README.

    NVRM: bad caching on address 0x100362d0000: actual 0x163 != expected 0x173
    NVRM: bad caching on address 0x100362d1000: actual 0x163 != expected 0x173
    NVRM: bad caching on address 0x1003be7c000: actual 0x163 != expected 0x173
    NVRM: bad caching on address 0x1003be7d000: actual 0x163 != expected 0x173

Please note: if you have checked the above items and continue to experience stability problems, please submit a bug report (see for information on how to submit bug reports).

zander is offline