Comment 113 for bug 148408

Some comments in the Debian bug report might indicate that it is a timing issue, and that anything slowing down the switch (like busy CPU, debug tracing, and perhaps accidentally doing a POST) works around it. The old patches mentioned here do not apply to the current code. Sparkling the code with microsleeps could work. I am not so optimistic about how much efforts the Xorg driver maintainers would invest in these old cards, but it might help if some of you having the card could do some testing. Try for instance attaching to Xorg with strace over the network from another computer. Or use gdb (also remotely) to narrow down where it hangs - for instance with breakpoints.