Bug 74062

Created attachment 92790 [details]
Video of a short gameplay in OpenArena where the artifacts can be seen

Created attachment 92791 [details]
Blender having artifacts visible only “inside” 3d elements in the design view

What hardware is this with? (dmesg | grep -i chipset). Do you see any nouveau-related errors in dmesg?

(In reply to comment #3)
> What hardware is this with? (dmesg | grep -i chipset). Do you see any
> nouveau-related errors in dmesg?

Hi Ilia, it is chipset GK104 (NVE4).
I don't see any nouveau related errors in dmesg, only a warning about unknown voltage. I am attaching the output of dmesg.

Created attachment 92889 [details]
Output of dmesg for the system with the issue.

I think I remember someone else having this style of corruption... e.g. https://bugs.freedesktop.org/attachment.cgi?id=87845 from bug #70326 (for nvf0, but they're sorta similar and the reporter mentions that he saw that issue on a gtx650, which is nve7 probably). Sadly, not the slightest clue as to what's wrong :( Feels like it's just not rendering certain triangles, or doesn't have time to do it, or we're not waiting for it properly... or something else entirely.

I was going to recommend that you try to use blob graph fw, but I see you're already doing that. I suppose going back to the nouveau one causes even worse issues...

(In reply to comment #6)
> I was going to recommend that you try to use blob graph fw, but I see you're
> already doing that. I suppose going back to the nouveau one causes even
> worse issues...

Actually, the nouveau one has exactly the same issues.

Probably useful, the artifacts are not present if the OpenGL application is run inside a Xephyr screen, though the performance decreases.

On the following link, you can find my experiences of this issue on nouveau with a GTX760 card: 
http://bugzilla.redhat.com/show_bug.cgi?id=1029224

Note that this bug only shows up when running high-end games on high graphics settings.
Other games, like Xonotic don't exhibit this problem (even on high quality) and run very smooth (60-120fps). Nouveau seems to use the lowest clock speed: my GPU idle temp is 32 degrees Celcius and climbs to only maximum 35 degrees Celcius on full load. (so I am quite pleased with nouveau: good performance for nearly no power consumption)

(In reply to comment #9)
> On the following link, you can find my experiences of this issue on nouveau
> with a GTX760 card: 
> http://bugzilla.redhat.com/show_bug.cgi?id=1029224
> 
> Note that this bug only shows up when running high-end games on high
> graphics settings.
> Other games, like Xonotic don't exhibit this problem (even on high quality)
> and run very smooth (60-120fps). Nouveau seems to use the lowest clock
> speed: my GPU idle temp is 32 degrees Celcius and climbs to only maximum 35
> degrees Celcius on full load. (so I am quite pleased with nouveau: good
> performance for nearly no power consumption)

My issue happens with Xonotic too. I dont know why.

Created attachment 100185 [details]
demonstrating more clearly the bug with glxgears

The bug being demonstrated with glxgears and resizing it to various window sizes so how the artifacts change can be seen. Note that the artifacts completely disappear when the scene rendering is smaller than a certain size.

Wild guess - perhaps it's LTCG related ?

You can easily check if that's the case my setting compressed = false in
nvc0_miptree.c:nvc0_miptree_create.

(In reply to comment #12)
> Wild guess - perhaps it's LTCG related ?
> 
> You can easily check if that's the case my setting compressed = false in
> nvc0_miptree.c:nvc0_miptree_create.

Seems like, I just did what you suggested and all artifacts are gone now (just had to restart any OpenGL program after installing). So I confirm this workaround works for me.
Thanks Emil.

It's worth retrying this with Linux 3.17.  I fixed a bug with the same symptoms on a macbook I had access to, but did not expect the same fix to be relevant for the other boards where the issue was reported.

However, I've just had an email saying it's fixed for another guy, so, it's worth a try.

(In reply to Ben Skeggs from comment #14)
> It's worth retrying this with Linux 3.17.  I fixed a bug with the same
> symptoms on a macbook I had access to, but did not expect the same fix to be
> relevant for the other boards where the issue was reported.
> 
> However, I've just had an email saying it's fixed for another guy, so, it's
> worth a try.

I just compiled Linux 3.17 and found that the bug won't happen with it.