I decided to give another chance to accelerated X on my Linux laptop today, as XGL has failed in the past. Apparently, AIXGL with Beryl (latest svn) worked easily, without major problems (just a few Beryl keyboard shortcut bugs). However, it’s painfully slow. Yes, it is hardware accelerated alright, it’s just that desktop actions don’t feel as instant as Mac OS X’s do. For example, in the 1-2 seconds that a fancy trick takes place, 0.5 sec of that time is far from smooth. It feels clunky. I run Arch Linux on a 2.8 Ghz P4 laptop with an ATi Radeon 9000 Mobility 64 MB (r250, using OSS Xorg “radeon” driver), while QuartzExtreme-accelerated Mac OS X’s animated tricks run way smoother on this 867 Mhz G4 with a 32MB GeForce4MX. I am betting that XGL is equally slow as AIXGL is (ATi’s binary driver does not support my r250-based chipset). From what I have seen so far, it’s dissapointing and I suspect that the problem is the architecture of the whole thing rather than the Ati driver. I am going back to plain X11.
Update: Oh, I forgot to mention the memory usage! Without accelerated X, my optimized Arch Linux uses 95 MBs of RAM on Gnome 2.16.2. With AIXGL, the default memory usage is about 200 MBs. The X server alone says that it consumes 98 Mbs of RAM. That’s pretty hefty IMHO for what it gives back to the user and it means that desktop distros like ubuntu/suse/fedora that normally require more memory than lightweight distros like Arch or Slackware, they would need 512 MBs PCs as minimum if they go accelerated by default (currently they are in the 256 MB area). Sure, they would work with 360 MBs of RAM too, but no such PCs are sold. Users usually have 256 MB PCs or 512 MBs or 1 GB. What I am saying is that accelerated X will have a huge impact in the minimum requirements of a popular Linux distro, and I am not talking about getting a good graphics card.