NVIDIA video cards, single display, SLI mode
From MEPIS Documentation Wiki
!!! Warning! The info contained in this article pertains to older versions of MEPIS !!!
This article discusses installing MEPIS 6.5 on a computer with dual CPU with two nVidia video cards and configuring nVidia Scalable Link Interface (SLI) mode for one monitor.
Initially, MEPIS 6.5 might not successfully install its copy of the official nVidia driver. Eventually you may need to install the official nVidia driver package. Before using nVidia's installer, make sure you uninistall any official nVidia driver and GLX you might have attempted through MEPIS' assistant, Synaptic, apt-get, or any other method.
Installing the official nVidia driver is beyond the scope of this article. But when installed and you see the nVidia logo, the driver is working but you must check /var/log/Xorg.0.log to be sure there are no problems preventing SLI from working.
Run this command in a shell:
cat /var/log/Xorg.0.log | grep "NVIDIA SLI"
and look for this:
(II) NVIDIA(0): NVIDIA SLI enabled.
If you find this, you are all set. Otherwise you might find problems; example: neither video card initializing properly. If so, there are two issues which must be addressed for SLI to work:
1) Too much RAM. If you have more than 1 GB RAM and running a 32-bit kernel, there is a problem during boot up that leaves no memory for the nVidia driver to initialize the video cards. The solution for this problem has several components:
a) Tell the kernel to reserve enough RAM during memory mapping. Add the following to the /boot/grub/menu.lst's kernel line:
320M is an example. 320M is required for the GeForce 8800 GTS which has 320 MB video RAM.
b) Disable frame buffer drivers to prevent mapping the the video card's memory into the kernel's virtual address space. Add the following to the /boot/grub/menu.lst's kernel line:
c) Tell the kernel there is only 512MB RAM in the system during bootup. This doesn't affect the system after booting is complete. Add the following line to /boot/grub/menu.lst just above the kernel line:
2) Multiple CPU, Multiple GPU, 32-bit kernel
Another problem occurs with systems containing multiple CPU (e.g. a dual core AMD Athlon 64 X2), multiple GPU's (e.g. two 8800 GTS) and a 32-bit kernel (e.g. linux-image-2.6.15-27-k7). Bugs in the Memory-Mapped PCI Configuration Space mechanism in older 2.6.x kernel cause problems. The solution is to turn this off by adding the following to the /boot/grub/menu.lst's kernel line:
Finally, here is an example grub entry:
title MEPIS at sda1, newest kernel root (hd0,0) uppermem 524288 kernel /boot/vmlinuz root=/dev/sda1 nomce quiet video=vesa:off vga=normal vmalloc=320M pci=nommconf resume=/dev/sda2 boot
NOTE: Your root line and your root= kernel parameter may be different. The parts described in this article are in bold.
Both these problems are detailed in the nVidia driver README. See "Kernel virtual address space exhaustion on the X86 platform" and "MMConfig-based PCI Configuration Space Accesses".