Testing request: owners of older GeForce cards (NV30 and NV40 families)

[AdamW]

Hi, folks. Time for some crowdsourcing!

So I'm trying to poke this rather infamous F16 / F17 bug:

https://bugzilla.redhat.com/show_bug.cgi?id=745202

What I'd like to get more info on is exactly what cards are affected by it.

The bug causes corrupted rendering of GNOME Shell elements on older NVIDIA GeForce cards, when using the nouveau driver. So far there seem to be two similar but slightly different forms of corruption discussed in the bug. The one that seems to be associated with NV30 cards looks like this:



And the one that seems to be associated with NV40 cards looks like this:



In both cases, there's clearly incorrect rendering of text in GNOME Shell elements like the top panel, the overview, and the User menu - but *not* in regular applications. In the NV30 case at least, the top panel is often an incorrect color (not black).

The cards under 'suspicion' are all those listed in the 'NV30 family' and 'NV40 family' here. Broadly, that's GeForce FX 5xxx, GeForce 6xxx cards and GeForce 7xxx cards, but it's a bit more complex than that - there's some Quadro (workstation cards) included. You can verify your exact card and compare it to the 'NVxx' model number given in that chart by looking either for a line "NOUVEAU(0): Chipset: "(foobar}" in /var/log/Xorg.0.log, or doing 'glxinfo | grep renderer' and looking at the result.

If you have any of the chips listed as being in the NV30 or NV40 families, can you please test an up-to-date F16 or F17 with gnome-shell - a live image of F17 is fine, the Alpha should be okay for testing this. PLEASE USE NOUVEAU. The bug is relevant to nouveau only. It does not happen with the proprietary driver, we know that, and we don't need reports from the proprietary driver. Also, the bug only affects Shell, so you have to test Shell - it doesn't happen in fallback mode, or KDE, or Xfce, or LXDE. Please post the following info:

* Your exact code name, as given in the Xorg.0.log or glxinfo | grep renderer output
* Whether you tried F16 or F17
* Whether you saw the corruption

If you see some other bug or some other form of corruption, that's interesting too, but please state that: it's important to know whether you actually hit the precise corruption discussed in 745202 (so, whether your screen looks like one of the two screenshots). If you're not sure, post a screenshot.

I'm particularly interested in reports of cards in the above two families which do *not* show corruption in Shell (but remember, USE NOUVEAU!), and reports from NV30, NV31, NV35 and NV36 cards, if anyone still has any of those.

Thanks a lot, everyone!

[glennzo]

Can't help with that one Adam. Don't have the required vid card.

[DBelton]

I think I have an NV41 card in my parts box that was working fine when I tossed it in there. I'll pull it out and see.

Only problem is, The one machine I have left that is AGP, I can't do an install on, so will running from the livecd be enough to test? I know a lot of times there are totally different results running from the livecd than from a full install.

Woohooo.. I also have a NV4B card, too. I think the fan is dead on that one, though. But I have a spare fan sitting here ready...

Actually, Adam.. That corruption looks like what has happened to me several times in the past when the fan on my video card died, and the GPU started getting a little hot. Is is possible that some of those reports are from people that the fans have gone out on their video card? Nvidia cards are notorious for bad fans. I have to replace just about every one of mine within the first couple of months after I get a new card.

Edit:
Another thought just hit me, too. It could possibly be a heat issue as a lot of those series cards did not even have fans on them. just a cheap heatsink.

But grabbing a livecd image now to run a few tests with here.

Just re-read your post and you are looking for both F16 and F17...

I have been running F16 here since the alpha with no corruption.

[Me@tower11 ~]$ glxinfo | grep render
direct rendering: Yes
OpenGL renderer string: Gallium 0.4 on NV4B

I will try that same machine, swapping out the card for the older in a few minutes.

Just another note... The card above has had issues like you are seeing in the past when the fan died on it, but I haven't seen anything like it since replacing the fan (that was way before F16, so I haven't seen the problem in F16)

[AdamW]

it's barely possible, but seems unlikely that so many reporters would have the problem due to heat.

iirc, as you say, many fx5200s had no fans; this is because they don't get hot enough to require them. i'd expect such cards to be less susceptible to heat issues, not more.

[DBelton]

I have noticed that using the nouveau drivers, my video cards seem to run a little hotter than with the nvidia drivers. Hasn't been enough of an issue for me to switch to the nvidia drivers since F15, though, because the nouveau drivers have worked so well for me here.

I just mentioned the heat issue because those screens look identical to what I saw when my video card got a little hot on me, but it wasn't hot enough for it to completely shut down. Just tossing in ideas in the hopes that something might jog a memory or shake loose a solution.

[StephenH]

I have a PNY GeForce 7600GS PCIe card, but no spare machine into which I can install it. The one which I would have used has an AGP slot instead.

[AdamW]

dbelton: nouveau's power management was, for a long time, non-existent, and now is still substantially inferior to NVIDIA's. that's why cards run hotter with nouveau than NVIDIA.

[RupertPupkin]

I get the same corruption as in the 2nd pic (letters blanked out) with a GeForce4 MX440 (NV17) in GNOME Shell in F16 using the nouveau driver. This is on my 2nd PC (not the one in my sig).

[Finalzone]

I used to own a Geforce 7800 GS a while ago. It now resides at the hands of Nouveau developers.

[robertdaleweir]

Hi, folks. Time for some crowdsourcing!

So I'm trying to poke this rather infamous F16 / F17 bug:

https://bugzilla.redhat.com/show_bug.cgi?id=745202

What I'd like to get more info on is exactly what cards are affected by it.

The bug causes corrupted rendering of GNOME Shell elements on older NVIDIA GeForce cards, when using the nouveau driver. So far there seem to be two similar but slightly different forms of corruption discussed in the bug. The one that seems to be associated with NV30 cards looks like this:



And the one that seems to be associated with NV40 cards looks like this:



In both cases, there's clearly incorrect rendering of text in GNOME Shell elements like the top panel, the overview, and the User menu - but *not* in regular applications. In the NV30 case at least, the top panel is often an incorrect color (not black).

The cards under 'suspicion' are all those listed in the 'NV30 family' and 'NV40 family' here. Broadly, that's GeForce FX 5xxx, GeForce 6xxx cards and GeForce 7xxx cards, but it's a bit more complex than that - there's some Quadro (workstation cards) included. You can verify your exact card and compare it to the 'NVxx' model number given in that chart by looking either for a line "NOUVEAU(0): Chipset: "(foobar}" in /var/log/Xorg.0.log, or doing 'glxinfo | grep renderer' and looking at the result.

If you have any of the chips listed as being in the NV30 or NV40 families, can you please test an up-to-date F16 or F17 with gnome-shell - a live image of F17 is fine, the Alpha should be okay for testing this. PLEASE USE NOUVEAU. The bug is relevant to nouveau only. It does not happen with the proprietary driver, we know that, and we don't need reports from the proprietary driver. Also, the bug only affects Shell, so you have to test Shell - it doesn't happen in fallback mode, or KDE, or Xfce, or LXDE. Please post the following info:

* Your exact code name, as given in the Xorg.0.log or glxinfo | grep renderer output
* Whether you tried F16 or F17
* Whether you saw the corruption

If you see some other bug or some other form of corruption, that's interesting too, but please state that: it's important to know whether you actually hit the precise corruption discussed in 745202 (so, whether your screen looks like one of the two screenshots). If you're not sure, post a screenshot.

I'm particularly interested in reports of cards in the above two families which do *not* show corruption in Shell (but remember, USE NOUVEAU!), and reports from NV30, NV31, NV35 and NV36 cards, if anyone still has any of those.

Thanks a lot, everyone!

Hi AdamW
I posted the following and it is still valid. I posted this at bugzilla a while back.
---->
Robert Weir 2012-02-09 17:25:32 EST

Hi
I was able to recreate this condition on a Desktop with a nVidia Corp nv34
[GEForce FX 5500] (rev a1).
Hopefully this can help with your endeavors...
Robert
<----
See the URL https://bugzilla.redhat.com/show_bug.cgi?id=767337 to see a couple of pictures etc.
Love to get this bug resolved, hope this helps...

[AdamW]

robert: yeah, I'm aware of that report.

rupert: that does reinforce the likelihood of the NV4x cases being VRAM exhaustion, thanks.

[JEO]

Hi Adam, I have a card that does *not* show the problem.
Using F17 Alpha Live Desktop, "OpenGL renderer string: Gallium 0.4 on NV4A" (this is a 6200 AGP VRAM: 256MB)

---------- Post added at 05:05 AM ---------- Previous post was at 04:33 AM ----------

I just tried it with another card in a different system that does *not* have the problem.

Still using F17 Alpha. "OpenGL renderer string: Gallium 0.4 on NV46" This is a 7300 GS PCIe. VRAM: 256MB

[bbfuller]

Using the F16 LiveCD

OpenGL renderer string: Gallium 0.4 on NV46

otherwise identified as nVidia Corporation G72 [GeForce 7300 LE]

Kernel driver in use: nouveau

256MB

No observable corruption to match either of the examples shown.

---------- Post added at 12:40 PM ---------- Previous post was at 12:28 PM ----------

Also, using the same Live CD

OpenGL renderer string: Gallium 0.4 on NV4C

nVidia Corporation C61 [GeForce 6150SE nForce 430]

Kernel driver in use: nouveau

256MB

Same results as previous, no observable corruption.

---------- Post added at 01:02 PM ---------- Previous post was at 12:40 PM ----------

On the other hand and using the same CD:

OpenGL renderer string: Gallium 0.4 on NV34

nVidia Corporation NV34 [GeForce FX 5200]

Kernel driver in use: nouveau

128MB

Exhibits the symptoms displayed in your second screenshot Adam with the added "attraction" of a rainbow hued top panel.

---------- Post added at 02:36 PM ---------- Previous post was at 01:02 PM ----------

And finally, I've got another:

OpenGL renderer string: Gallium 0.4 on NV34

nVidia Corporation NV34 [GeForce FX 5200]

Kernel driver in use: nouveau

64MB

probably from another manufacturer that displays the top panel in a different way but otherwise fails in the style of your second screen shot Adam.

The difference with the top panel is that at first it didn't appear at all. After the screen blanker had set in I waggled the mouse to get the display back and got some fairly violent display flashing before the desktop returned. There's still no actual top panel, but a few isolated, corrupted bits here and there. A bit like shot 2 but no discernible background.

[StephenH]

I used to own a Geforce 7800 GS a while ago. It now resides at the hands of Nouveau developers.

I would be willing for my old GeForce 7600 GS to go the same route. My primary machine has a GeForce GT220 currently. I don't see a need for the old 7600GS card any time soon.

[AdamW]

Thanks for the reports folks! Very helpful. Certainly seems to establish that not all NV4x cards are suffering from the bug. All NV34s do seem to have it.

BTW, one more thing it would be useful to know for NV4x reports (whether you see the corruption or not) is how much video RAM the card has (not system RAM, but the memory on the video card). We suspect the NV4x cards that have the problem are suffering from VRAM exhaustion (they just don't have enough VRAM).