Originally Posted by netllama
What you're experiencing is the driver detecting a specific class of errors and attempting to recover from them. This is a new feature in 1.0-9625 which was not present in the 1.0-8xxx (and earlier) drivers. This is why the problem is much worse for you in 1.0-8774.
The official FC kernels have AGP support compiled in, which is why you cannot use the nvidia driver's internal AGP support.
If you set NvAGP to 0, does that have any impact on the instability?
Thanks for the quick answer.
I suspected the new feature, which is pretty interesting.
I would be curious to know which "class" is this one... :-)
About the kernel, it's not an official FC, but a vanilla 2.6.18 with agpgart compiled as module. What I suspect is that the module (probably the intel_agp) is loaded by udev, I was just wondering if there is a way to avoid this (or if something else is the cause).
I'll try NvAGP 0 and let you know.