nVidia driver crash bug: Please check your logs!

[loftar]

As some will no doubt already be aware, there is an elusive bug in conjunction with nVidia graphics cards that causes a crash to desktop. This bug has been difficult to track down, as for some reason it seems to only show up in somewhat large PvP scenarios, and neither I nor anyone else seems to have been able to come up with any method for reproducing it in a controlled manner. The fact that the crash happens in the driver also makes it difficult to collect ancillary information. Because of all this, this bug has remained elusive for well over a year at this point, despite it being quite critical.

Some time ago, someone (I only heard of it indirectly, so I actually don't know who) managed to figure out a way to make the driver crash in a seemingly identical manner. It should be well noted that crashing in the same manner doesn't necessarily mean that it crashes for the same root cause, but nevertheless it's the best (or only, really) lead that I've got so far. Specifically, the crash happens when issuing an OpenGL draw command that takes data from a zero-sized Vertex Buffer or Index Buffer. Since a while back, therefore, I've made the client emit warnings whenever it attempts to create such a buffer object, however the only mechanic I've been able to find that has done that has been the falling leaves in autumn (and it also avoids the actual crash by not actually attempting to draw any vertices), and that's clearly not the cause of the issue, since crashes haven't only happened during autumn.

Therefore, I'd appreciate it if you all could check your warning logs, especially any of you who have experienced a crash-to-desktop lately (but regardless, really; it's not like it's guaranteed to cause a crash). These critical warnings are saved in your user home directory (which, on Windows, is usually C:\Users\<your-username>\), and is called haven-errors.log. If you have a file of that name in your home directory, please convey its contents to me in one way or another, by posting it directly in this thread if it's small enough, using the forum Attachment function if it's larger than that, or in some other manner of your choosing.

Of course, I don't know how many custom clients have actually merged this warning functionality, and it would be unfortunate if a useful warning had been lost, but let's hope someone has logged a clue. It would be quite great if this bug could finally be fixed!

[Clemins]

as for some reason it seems to only show up in somewhat large PvP scenarios, and neither I nor anyone else seems to have been able to come up with any method for reproducing it in a controlled manner. The fact that the crash happens in the driver also makes it difficult to collect ancillary information. Because of all this, this bug has remained elusive for well over a year at this point, despite it being quite critical.

Glue together a small test server with natural world gen, invite people who are familiar with fighting, and lets do some tests.

[DonVelD]

I unfortunately do not have any log to share, although I'd like to ask about the idea with gathering as many players as possible in a combat scenario and trying to reproduce the bug. I don't think anyone would actually bother to do it by themselves, and mass PvP is nonexistent at this point of the world - so it'd be the best if you encouraged it with perhaps a stream. Do you have something like this planned yet?

[Nightdawg]

Move this thread to announcements and pin it.

[loftar]

Just using the natural world-gen seems a tad naïve, as the cause of the error could just as well be parts of the constructed environment, mounts, equipment, or even something completely unexpected, something that may quite conceivably be better explored in eg. Valhalla than in a from-scratch world.

[loftar]

It should perhaps also be added that there is a potential workaround in simply changing the client to force a certain minimum size of buffer objects, and if this drags out too much further, I do plan on implementing that workaround, but I'd really like to find the actual root cause, if at all possible. (Especially given that, as mentioned, it's not even guaranteed that the buffer-size problem is the actual cause of the crash.)

[DonVelD]

Just using the natural world-gen seems a tad naïve, as the cause of the error could just as well be parts of the constructed environment, mounts, equipment, or even something completely unexpected, something that may quite conceivably be better explored in eg. Valhalla than in a from-scratch world.

Then maybe you could copy a large chunk of the world and try to work with that? It'd bring more people than trying it in the real game at least, because of travel etc. We also could try all of em, Valhalla, natural world-gen, and copied part of the world. If nothing would work then I guess the last thing that could be tried would be the real thing.

Doing something like that at least is better than just waiting and asking for people to find the bug.

[Clemins]

Just using the natural world-gen seems a tad naïve, as the cause of the error could just as well be parts of the constructed environment, mounts, equipment, or even something completely unexpected, something that may quite conceivably be better explored in eg. Valhalla than in a from-scratch world.

Well, I figured since world generated objects weren't fully ruled out, including it in a test environment wouldn't be a bad idea. If you suspect the bug might be from a constructed environment and want to use Valhalla as a test zone, perhaps you could either place a bunch of player made structures and items around or allow testers to build objects w/o the required building materials. Either way, I am willing to help assist getting this bug figured out in a test environment.

[mulamishne]

Haven and Hearth, AMD sponsorship confirmed. I could deal with a scheduled testing shit fest whenever. Maybe do it as part of a loftar devstream?

[loftar]

Maybe do it as part of a loftar devstream?

I don't have any streaming equipment, so I think it would at least have to happen on Jorb's terms. :)