Comment 7 for bug 368187

Further to the above tests, I wanted to check if Karmic would solve the issue but it's actually even worse.
I tried the live CD of Karmic Alpha5 for AMD64 and not only the screen stays black, but there seems to be an init sequence repeating over and over (from observing cursor changes) and if you let a few sequences run, Ctrl+Alt+F1 is no longer responsive. On the second try, I pressed quicker and was able to copy the logs (in attachment).

A difference of interest with Karmic is this one additional dmesg line that I noticed since it also appeared on screen just before the black screen:
[ 0.120997] pci 0000:00:00.0: BAR 0: address space collision on of device [0xe0000000-0xe3ffffff]
This same address space is mentioned towards the end when drm fails:
[ 96.398561] [drm] Can't use AGP base @0xe0000000, won't fit

Then, I took my old hardy CD (AMD64), did the same try and everything worked (logs attached as well). So I don't know what differences were introduced since hardy but it could be worth looking into the diffs for extra information.
For the sake of comparison, I captured the ressources tabs of WinXP SP3 device manager, where I saw that the particular [0xe0000000-0xe3ffffff] was used by three devices: the HD3650 AGP, CPU to AGP Ctlr and PCI bus.

I'm not sure what else I can do to give this bug more importance. One thing for sure is with my hardware, I cannot do fresh install of either jaunty or karmic, so this is a pretty severe bug for me, and since it worked with hardy, it's actually a regression. I'm no expert but since the address space collision happens this early in the log, isn't it more a kernel bug than a ati driver bug?

Let me know if I can help...