Currently we have 3 gfxcard-intel* mix-ins:

1. gfxcard-intel, which defaults to mean the same as gfxcard-intel-classic on 1.4 release, and gfxcard-intel-iris on the next release.
2. gfxcard-intel-classic, which will set VIDEO_CARDS to "intel".
3. gfxcard-intel-iris, which will set VIDEO_CARDS to "intel i965".

Basically, the difference is that if the user chooses the "classic" one, they will get the non-kms i915 mesa driver from the (x11-drivers/xf86-video-intel ebuild). Whereas, if the "iris" one is chosen, then the i965 or iris modesetting enabled driver integrated in the xorg-server will be used instead.

Users with a 3rd generation or older chipsets need to choose the classic one. Other users should choose the iris one.

Both gfxcard-intel-classic and gfxcard-intel-iris will enable the "xa", "xvmc", "vaapi" and "vdpau" USE flags, although not every intel video card meant to be represented by the mix-ins supports every one of those features.

Also, users with a video card that supports the VAAPI will not be done by just enabling the corresponding mix-in.  They will have to manually install media-libs/libva-intel-driver or media-libs/libva-intel-media-driver depending on the video card they have.

We have actually 3 groups of intel graphics, according to the feature set they support, which could be configured as follows:

1. Needs the i915 mesa driver and doesn't support VAAPI or Vulkan.

   VIDEO_CARDS="intel i915"

1. Should use the i965/iris driver and supports VAAPI but not Vulkan.

   VIDEO_CARDS="intel i965"
   USE="xa xvmc vaapi vdpau"
   block x11-drivers/xf86-video-intel
   install media-libs/libva-intel-driver
   

1. Should use the i965/iris driver and supports both VAAPI and Vulkan.

   VIDEO_CARDS="intel i965"
   USE="xa xvmc vaapi vdpau vulkan"
   block x11-drivers/xf86-video-intel
   install media-libs/libva-intel-media-driver
   

These could be called:

1. gfxcard-intel-classic
2. gfxcard-intel-4thgen
3. gfxcard-intel-8thgen

If a generic gfxcard-intel mix-in should exist, then its definition should be tied to the sub-architecture, which matches better the probable chipset generation (and therefore feature set) instead of the Funtoo release as it is now.