The nVidia drivers include kernel modules that must integrate in your current kernel. To accomplish this, your kernel must support the loading of kernel modules. If you used genkernel to configure the kernel for you then you're all set. If not, double check your kernel configuration so that this support is enabled:

Loadable module support --->
  [*] Enable loadable module support

You also need to enable Memory Type Range Register in your kernel:

Processor and Features --->
  [*] MTRR (Memory Type Range Register) support

nVidia's modules and libraries are combined in two packages: nvidia-glx and nvidia-kernel. The former are the X11 GLX libraries while the latter are the kernel modules. Since nvidia-glx depends on nvidia-kernel, installing nvidia-glx is sufficient:

# emerge nvidia-glx

Once the installation has finished, run modprobe nvidia to load the kernel module into memory.

# modprobe nvidia

You probably want to have this done each time you boot your system, so edit /etc/modules.autoload.d/kernel-2.6 (or kernel-2.4) and add nvidia to it. Don't forget to run modules-update afterwards!

# modules-update

Once the appropriate drivers are installed you need to configure your X Server (XFree86 or Xorg) to use the nvidia driver instead of the default nv driver.

Open /etc/X11/XF86Config (or /etc/X11/xorg.conf) with your favorite editor (such as nano or vim) and go to the Device section. In that section, change the Driver line:

Section "Device"
  Identifier "nVidia Inc. GeForce2"
  Driver     "nvidia"
  VideoRam   65536
EndSection

Then go to the Module section and make sure the glx module gets loaded while the dri module doesn't:

Section "Module"
  (...)
  # Load  "dri"
  Load  "glx"
  (...)
EndSection

Run opengl-update so that the X Server uses the nVidia GLX libraries:

# opengl-update nvidia

To test your nVidia card, fire up X and run the glxinfo | grep direct command. It should say that direct rendering is activated:

$ glxinfo | grep direct
direct rendering: Yes

To monitor your FPS, run glxgears.

If you are having troubles with the nVidia 2D acceleration it is likely that you are unable to set up a write-combining range with MTRR. To verify, check the contents of /proc/mtrr:

# cat /proc/mtrr

Every line should contain "write-back" or "write-combining". If you see a line with "uncachable" in it you will need to change a BIOS setting to fix this.

Reboot and enter the BIOS, then find the MTRR settings (probably under "CPU Settings"). Change the setting from "continuous" to "discrete" and boot back into Linux. You will now find out that there is no "uncachable" entry anymore and 2D acceleration now works without any glitches.

The current nvidia-kernel package only supports kernels using an 8K stack size. More recent kernels (2.6.6 and higher) have support for 4K stack size's as well. Do not select 4K stack size in your kernel configuration. You can find this option in the section Kernel Hacking.