nVIDIA Hateage

nVIDIA has released the first driver in the 8xxx series for their video chipsets. The release notes look pretty sparse for a major build number jump. I was told that the issue with X eating 100% CPU would be fixed in this release, but it looks like it isn't. I'm getting the same "Xid" messages in syslog, along with "NV(0): WAIT (.....)" messages in the X server log which I don't believe had been appearing before. Unlike before, I can't save the X session by killing and restarting xfwm4; I have to bring the entire X process down remotely.

I've disabled window drop shadows for the time being to see if that helps solve the problem. Update: It doesn't.

I'm really getting sick and tired of this. nVIDIA appears incapable of releasing a stable driver set. I understand that X.org 7.0 hasn't been released yet, and Composite is new and experimental, but many other chipsets (such as the ATI Radeon Mobility 7500 in my laptop) work nice and stably with Composite and Exa enabled in the 7.0 release candidate. Just another data point in the huge mess of reasons why OSS is better than closed-source proprietary shit. I tried the OSS nv driver earlier, but simply moving a translucent window across half the screen took about 5 seconds. Not pleasant. I tried nv's experimental Exa support, but it crashes the X server on startup. Oh well. If I knew more, I'd try to hack on it.

I'm seriously considering tossing this card and buying an ATI instead. At least they release partial specs so OSS devs can write accelerated 2D drivers, even if the 3D acceleration is still a bit of a mystery. If my little 4-year-old laptop can handle Composite, I'm sure a newer board can as well.

Alternatively, anybody know of a decent video chipset with stable OSS drivers with acceleration that can handle Composite, without breaking the bank?

Update: As ElAngelo points out, nvidia released a quick fix-up release to hopefully fix the problems I'm experiencing. So far so good. On a side note, with the old drivers, it seems that disabling fast writes and sideband addressing seems to help the problem quite a bit: I didn't get a lockup for a couple days, vs. the usual couple hours.