I don't know ---I think so.
It really seems like, if there is supposed to be support for KT333 under the umbrella of Nvidia KT266 support, then it's not really working for all kinds of KT333 boards. (A kt333 chipset is in many ways similar to kt266. It has the same pinout or whatever, and i think in some aspects beyond the agp interface, kt333 support is really an extension of kt266 support... But it doesn't seem that linux or nvidia support for kt333 is completely solid. There are some quirks that aren't being caught )
If the Nvidia driver is supposed to work under Linux's generic gart driver, then it's not stable on all Kt333's . My GF2 works with that combination, ( nvidia glx / agpgart) but my GF4 does not (for more than about 15 seconds).
If there was a problem with the motherboard and agpgart, then my GF2 should be unstable too. But it's not. Both cards have the same 4x agp bandwidth. If there was a big problem with my Gainward ti-4200 then it should not have worked OK in windows. But it did.
If it's a question of drive strength, (say the GF4 requires much more juice than GF2) well I have not encountered any guide for dealing with that. You aren't messing with agp drive strength and your Ti-4200 works fine in W2K.
The Asus A7V333 was designed to be manufactured with an AGP-Pro option. Seems like the engineers who did that work would have also been sensitive to questions of having enough current to drive non-AGP-pro cards as well on the normal AGP models.
You have a top quality powersupply and so do I. We both bought 50 watts over what is usually considered "Plenty". It might be an electrical problem but that just seems ruled out: the Asus engineers would have to have just completely messed up. I doubt that. The reason I go further and deny it as well as doubting it: W2K worked fine with your GF4, and also because my GF4 card crashed exactly the same way that I'm used to (Algae-vision) on a 300 watt SiS735 based Mandrake system using NvAGP 3. (and that chipset is supposedly supported- so it was definitely using the Nvidia gart). However, it did not crash on the same SiS735 system using the ever-so stable Windows98.
Both systems would have to be electrically flawed-- but only under Linux, not Windows.
The first half of that proposition is improbable and second half makes no sense; so it is not an electrical problem.
Is it a problem with interrupts? Not in the basic sense. I have the nvidia card on an irq all by itself and I can see it gets irq11... The GF2 also uses irq11 and does not crash. There is some question in my mind about the VIA irq router which does not "take" irq 11 in a persistent way, but I think does use it. I have no idea what it really does besides probably reassigning interrupts to devices and sharing them when needed. The same motherboard settings seem stable on SiS735 and your AsusA7V333 using Windows. The same BIOS settings and kernel on my Asus are stable using a GForce2 .
WHat can I conclude? There is a problem with IRQ routing in Linux kernel 2.4.20-ac2, but that it only shows up using a GF4 not a GF2 ? And when used with a different kernel, on a different chipset (non VIA board) produces the same kind of crash with the Gforce4 ....
Irq routing in Linux doesn't seem to be the problem, either.
Then there is the whole issue of the GF4 appearing to use agpgart more or less stably when the opengl applications are running in windows (stably compared to my fullscreen crashes which happen in mere seconds).
If there was something screwey at the hardware or BIOS level wouldn't glxgears just crash the system too, like fullscreen tuxracer or Q3 ?
The constant thread in all those variables is the Nvidia device driver, the TI-4200 and crashing Linux. Looking at the Gainward Windows installation software, basically it just installs Nvidia's Detonator drivers without even renaming/rebranding them as "Gainward device driver software". I haven't confirmed that this Detonator driver makes the GF4 stable on my Asus board ( I'd have to defile my own system with a Windose installation!) but it sure looks like the problem is that the Nvidia Unifried Driver works well in Windows, but not so swell in Linux.
Last edited by junkieclown; 01-18-03 at 04:08 PM.