GPU dead out of nowhere? Weird symptoms.

[Dr. Dro]

So it happened again. This time I went straight to power off the monitor, plug off the cable, wait 3 minutes and plug it in again. Voilá, instantly works again.

I hate hardware sometimes, I love it another times. This time it skews towards hate

It is possible that it's a symptom of pre-failure, keeping an eye on that atikmdag_dce.log file to see at which time it disconnected and what EDID data is being transferred at the time of the failure can help pinpoint it... but it is indeed quite odd.

 

[Kodehawa]

It is possible that it's a symptom of pre-failure, keeping an eye on that atikmdag_dce.log file to see at which time it disconnected and what EDID data is being transferred at the time of the failure can help pinpoint it... but it is indeed quite odd.

The file doesn't generate for me on C:/AMD, at least.

EDIT: oh, it's in C:/Program Files/AMD

The closest I could find to the link failing is

[00:05:02:25:239:355] [Conn_Link_Status] [DP-1] link is :On, [0 link training failed/1 attempted]
[00:05:02:25:245:139] [Conn_OS_TF_CS] - TF[sRGB] - CS[sRGB]
[00:05:02:25:245:150] [Conn_Pixel_Format] - [ARGB8888]
[00:05:07:20:503:439] [Conn_Mode] [DP-1] {2560x1440, 2560x1440@165:0, 2640x1470@639740Khz}
[00:05:07:20:503:464] [Conn_Freesync] [DP-1] Freesync E:0 SS:0 F:0
[00:05:47:44:728:530] [Conn_Link_Status] [DP-1] link is :Off, [0 link loss/232 minutes]
[00:05:47:44:728:545] [Surface_Formats] [DP-1]
Layer 0: RGB8:1 RGB10:0 FP16:0 NV12:0 P010:0
Layer 1: RGB8:1 RGB10:0 FP16:0 NV12:1 P010:0
Layer 2: RGB8:0 RGB10:0 FP16:0 NV12:1 P010:0
[00:06:23:46:660:455] [Conn_Mode] [DP-1] {2560x1440, 2560x1440@165:0, 2640x1470@639740Khz}
[00:06:23:46:660:495] [Conn_Freesync] [DP-1] Freesync E:0 SS:0 F:0
[00:06:23:46:796:133] [Conn_Mode] [DP-1] {2560x1440, 2560x1440@165:0, 2640x1470@639740Khz}
[00:06:23:46:796:164] [Conn_Freesync] [DP-1] Freesync E:0 SS:0 F:0
[00:06:23:46:840:120] [Conn_LKTN] [DP-1] HBR2x4 pass VS=3, PE=0
[00:06:23:46:840:132] [Conn_Link_Status] [DP-1] link is :On, [0 link training failed/1 attempted]
[00:06:23:46:845:808] [Conn_OS_TF_CS] - TF[sRGB] - CS[sRGB]
[00:06:23:46:845:821] [Conn_Pixel_Format] - [ARGB8888]
[00:06:23:46:847:117] [Conn_OS_TF_CS] - TF[sRGB] - CS[sRGB]
[00:06:23:46:847:122] [Conn_Pixel_Format] - [ARGB8888]
[00:06:23:47:472:000] [Conn_Link_Status] [DP-1] link is :Off, [0 link loss/268 minutes]
[00:06:23:47:744:418] [Conn_Link_Status] [DP-1] link is :On, [0 link training failed/1 attempted]
[00:06:23:47:792:990] [Conn_Hotplug] [DP-1] - Connected
[00:06:23:47:793:001] [Conn_Link_Status] [DP-1] link is :On, [0 link training failed/1 attempted]

The file is too massive to find anything on the sea of text, though.

Here's an EDID block:

[00:06:54:47:663:496] [Conn_Link_Status] [DP-1] link is :On, [0 link training failed/1 attempted]
[00:06:54:47:669:564] [Conn_OS_TF_CS] - TF[sRGB] - CS[sRGB]
[00:06:54:47:669:577] [Conn_Pixel_Format] - [ARGB8888]
[00:06:54:47:736:491] [Conn_OS_TF_CS] - TF[sRGB] - CS[sRGB]
[00:06:54:47:736:494] [Conn_Pixel_Format] - [ARGB8888]
[00:06:54:58:392:661] [Conn_Link_Status] [DP-1] link is :Off, [0 link loss/0 minutes]
[00:06:54:58:427:460] [Conn_Hotplug] [DP-1] - Disconnected
[00:06:54:58:427:477] [Surface_Formats] [DP-1]
Layer 0: RGB8:1 RGB10:0 FP16:0 NV12:0 P010:0
Layer 1: RGB8:1 RGB10:0 FP16:0 NV12:0 P010:0
Layer 2: RGB8:0 RGB10:0 FP16:0 NV12:0 P010:0
[00:06:54:58:440:057] [Conn_Mode] [DP-1] {2560x1440, 2560x1440@165:0, 2720x1559@699500Khz}
[00:07:07:27:338:443] [Conn_Link_Status] [DP-1] link is :On, [0 link training failed/1 attempted]
[00:07:07:27:448:320] [Conn_Color_Depth_Caps] [DP-1] - Supported Color Depths: 10 8 6 bpc
[00:07:07:27:448:331] [Conn_Display_Brightness_Caps] [DP-1] - Min[492] - Avg[322] - Max[408]
[00:07:07:27:448:340] [Conn_Disp_Cap_HDR_Metadata_Colorimetry] - Area[1455689830] - R[67675,31152] - G[27539,66796] - B[15136,5273] - WP[31347,32910]
[00:07:07:27:448:348] [Conn_Edid_Block_Type] [DP-1] EDID blocks: 2 Edid14Base Ext_CEA
[00:07:07:27:448:378] [Conn_Detect] [DP-1] G27q-20: [EDID Block 0]

00 FF FF FF FF FF FF 00 30 AE C3 66 42 47 5A 31
17 1F 01 04 B5 3C 22 78 3B 78 E5 AD 4F 46 AB 26
0D 50 54 BD EF 00 81 80 95 00 B3 00 D1 C0 61 68
61 7C 45 68 45 7C 56 5E 00 A0 A0 A0 29 50 30 20
35 00 55 50 21 00 00 1E 00 00 00 FF 00 55 36 33
33 31 5A 47 42 0A 20 20 20 20 00 00 00 FD 00 30
A5 FA FA 41 01 0A 20 20 20 20 20 20 00 00 00 FC
00 47 32 37 71 2D 32 30 0A 20 20 20 20 20 01 02
02 03 32 F1 49 01 02 03 04 11 12 13 90 1F 23 09
07 07 83 01 00 00 E2 00 D5 6D 1A 00 00 02 01 30
A5 00 00 00 00 00 00 E3 05 C0 00 E6 06 07 01 61
56 1C 14 EC 00 A0 A0 A0 67 50 20 30 35 00 55 50
21 00 00 1A 70 A0 00 A0 A0 A0 46 50 30 20 35 00
55 50 21 00 00 1E 70 C2 00 A0 A0 A0 55 50 30 20
35 00 55 50 21 00 00 1A E6 F9 00 50 A0 A0 1E 50
08 20 08 04 55 50 21 00 00 1A 00 00 00 00 00 65

 

[caroline!]

I had the same issue with my HDMI to VGA converter where it would randomly display that black screen or green/pink artifacting and fixed it by disabling HDMI Link Assurance and HDCP Support, and enabling GPU scaling which was off by default.

That did it for me.

 

[Dr. Dro]

I had the same issue with my HDMI to VGA converter where it would randomly display that black screen or green/pink artifacting and fixed it by disabling HDMI Link Assurance and HDCP Support, and enabling GPU scaling which was off by default.

That did it for me.

However, disabling HDCP support will also break certain DRM-protected media, i.e. you can't watch movies on your PC anymore. I don't recommend doing that for that reason alone.

A cable replacement (presumably a higher-quality cable with better shielding) improved the situation, so it is still quite possible that the problem is electrical in nature.

 

[R-T-B]

However, disabling HDCP support will also break certain DRM-protected media, i.e. you can't watch movies on your PC anymore. I don't recommend doing that for that reason alone.

A cable replacement (presumably a higher-quality cable with better shielding) improved the situation, so it is still quite possible that the problem is electrical in nature.

Generally that's true. But in Carolines case, you really aren't going to get HDCP across a VGA link anyways, so she should be fine.

 

[Kodehawa]

It is possible that it's a symptom of pre-failure

The most confusing part is that plugging in the HDMI port works. It's literally just something to do with DP going haywire on the monitor side lmao

 

[LabRat 891]

Whoa.
I knew cables can be 'strange' causes for problems, but this is a new one for me.

As I was reading OP, I was certain his card was bad (from similar experiences, years-past). When I got reading how it worked in Linux, I was confused...
Thank you for editing OP with the solution; just added something new to my 'toolbox'.

Edit: if the issue still comes up occasionally, but the new cable largely improved things, I'd recommended investing in a particularly well shielded DP cable with 'chokes/beads' on both ends.
'Spooky at a distance' (RFI/EMI) issues seem to be extremely common these days.
Eh, so much for 'FCC Part 15'

 

[Kodehawa]

Whoa.
I knew cables can be 'strange' causes for problems, but this is a new one for me.

As I was reading OP, I was certain his card was bad (from similar experiences, years-past). When I got reading how it worked in Linux, I was confused...
Thank you for editing OP with the solution; just added something new to my 'toolbox'.

Strange would be an understatement right now lmao
After 3 months or so of not shutting down the monitor, it (presumably) starts sending corrupt EDID data to the driver, which causes the weirdness you see. Two different DP cables do the same when it happens.
... On HDMI on my macbook works, on linux WITH DP it works.

Normally power cycling the monitor does NOT work. I have to shut it down, plug the cable out and wait several minutes. After plugging it on again, it powers on and works just fine.

Geez.

 

[LabRat 891]

Strange would be an understatement right now lmao
After 3 months or so of not shutting down the monitor, it (presumably) starts sending corrupt EDID data to the driver, which causes the weirdness you see. Two different DP cables do the same when it happens.
... On HDMI on my macbook works, on linux WITH DP it works.

Normally power cycling the monitor does NOT work. I have to shut it down, plug the cable out and wait several minutes. After plugging it on again, it powers on and works just fine.

Geez.

Sounds like the monitor has bad 'filtering circuitry'. I know on PCIe*, termination resistance and capacitance is very important, and poorly-engineered or damaged components can cause 'issues'.

Ex. I know of at least 2 instances of 'cost-engineered' mining Polaris GPUs having all sorts of odd issues come up when in (higher signal-rate and frequency) PCIE gen3. Switching to the lower frequency Gen2, fixes the issues.

*(IIRC) PCI-e and DP (and Ethernet) use conceptually-similar Packetized Serial I/O schemes.
I'm making 'a leap of logic' that Similar Failure-Symptoms may also have Similar Failure-Modes.

The most confusing part is that plugging in the HDMI port works. It's literally just something to do with DP going haywire on the monitor side lmao

Have you looked thru your monitor's OSD? Reset to defaults? Checked if VRR/FreeSync/Vsync are enabled/disabled in the OSD?
I could imagine a particularly poor implementation of VESA VRR (or 'overdrive') could corrupt or misinterpret data.
I'm not actually aware of how VESA VRR, etc. 'Co-signal' frame synchronization.

Also, as silly as it sounds, have you tried another power source for your monitor?

I've had a failing/failed PSU in an HDTV cause some whacky problems (and half-sine AC on the ground plane).
Might want to take a multi meter to any exposed metal or the ground shields on the DP/HDMI ports and 'actual earth/ground'. Stray AC or pulsed DC where it doesn't belong can surprisingly 'cause problems' more complex than merely "letting the magic smoke out".

(Also, I've ran across several users here and elsewhere over-the-years with 'issues' dependant on what exactly else was 'interacting' with the Branch Circuit their device was plugged into.)

 

[Kodehawa]

Sounds like the monitor has bad 'filtering circuitry'. I know on PCIe*, termination resistance and capacitance is very important, and poorly-engineered or damaged components can cause 'issues'.

Ex. I know of at least 2 instances of 'cost-engineered' mining Polaris GPUs having all sorts of odd issues come up when in (higher signal-rate and frequency) PCIE gen3. Switching to the lower frequency Gen2, fixes the issues.

*(IIRC) PCI-e and DP (and Ethernet) use conceptually-similar Packetized Serial I/O schemes.
I'm making 'a leap of logic' that Similar Failure-Symptoms may also have Similar Failure-Modes.


Have you looked thru your monitor's OSD? Reset to defaults? Checked if VRR/FreeSync/Vsync are enabled/disabled in the OSD?
I could imagine a particularly poor implementation of VESA VRR (or 'overdrive') could corrupt or misinterpret data.
I'm not actually aware of how VESA VRR, etc. 'Co-signal' frame synchronization.

Also, as silly as it sounds, have you tried another power source for your monitor?

I've had a failing/failed PSU in an HDTV cause some whacky problems (and half-sine AC on the ground plane).
Might want to take a multi meter to any exposed metal or the ground shields on the DP/HDMI ports and 'actual earth/ground'. Stray AC or pulsed DC where it doesn't belong can surprisingly 'cause problems' more complex than merely "letting the magic smoke out".

(Also, I've ran across several users here and elsewhere over-the-years with 'issues' dependant on what exactly else was 'interacting' with the Branch Circuit their device was plugged into.)

I did try freesync disabled (now) and enabled (last time it happened). No change

Cannot try another power source, there's a single source going through my room. Would be quite a pain, though.

Sadly (?) the PSU on this monitor is internal. Hopefully its not failing. I'll get a multi-meter soon and check once it happens again

 

[eidairaman1]

Whoa.
I knew cables can be 'strange' causes for problems, but this is a new one for me.

As I was reading OP, I was certain his card was bad (from similar experiences, years-past). When I got reading how it worked in Linux, I was confused...
Thank you for editing OP with the solution; just added something new to my 'toolbox'.

Edit: if the issue still comes up occasionally, but the new cable largely improved things, I'd recommended investing in a particularly well shielded DP cable with 'chokes/beads' on both ends.
'Spooky at a distance' (RFI/EMI) issues seem to be extremely common these days.
Eh, so much for 'FCC Part 15'

Im using a 10 foot dp cable on a 21" NEC monitor on a R7 250X.

 

[LabRat 891]

Stupid inquiry:
Have you given the DP port on your display "an NES Blow Job" yet?
IIRC, Resistance is part of signal termination; merely dust or corrosion could potentially be involved (no pun intended).

Im using a 10 foot dp cable on a 21" NEC monitor on a R7 250X.

Not 100% sure of the 'context' here, but I'll work off my first assumption(s)

Not everyone will have 'interference' issues, even using the exact same (or 'worse') cables/hardware.

Something like a nearby industrial arc furnace, poorly routed in-building HVAC power, or a neighboring hobbyist saying 'F*ck you' to FCC regulations easily can introduce interference well beyond what is engineered to be tolerable. Merely a 'defectively constructed cable' amongst thousands of units produced can also become an issue, even 'in normal cirumstances'.

Ex. At least 2x, I've ran across out-of-spec HDMI cables that were liable to induct near-ish-by ESD events; the link would crash, or the entire machine would crash. (One instance was a 'defective cable' in a customer's home; the other was an entire mass-purchase of 'cheap AF' HDMI cables by the IT dept. of company I was working at.)

All that said, in OP's case this is looking like an issue w/ the monitor itself.
Cable changes 'helped' but didn't resolve the issue; and IIRC, even other non-PC PCs can have a problem. The fact HDMI 'works over time' and DP doesn't, would imply to me either a firmware issue in how each link is processed/handled, or that the DP port on the display has a defect somewhere in its circuits.
(AFAIK, it's not uncommon for HDMI to 'default' to post-processing the input; while DVI, DP, and VGA inputs typically 'reproduce the image "raw"'.)

 

[delshay]

I did try freesync disabled (now) and enabled (last time it happened). No change

Cannot try another power source, there's a single source going through my room. Would be quite a pain, though.

Sadly (?) the PSU on this monitor is internal. Hopefully its not failing. I'll get a multi-meter soon and check once it happens again

I had this issue with my Vega card. It was fixed with high end HDMI cable, but you don't have to do this. I think you need a better quality cable preferably a "slightly shorter cable" which mine is which fixed the fault, never to return. I just think the odd card are sensitive to this & I know one of my cards are. What I was getting was random black screen, dropout.

I think you need a slightly shorter HDMI or Display port cable.

EDIT: Just for reference I went from 1.5M to just under a meter, but you don't have to go this length, just go shorter that you already have.

EDIT2: I'v just gone back to old cable. This is to investigate if it's a cable fault or a card fault. ...I think this may turn out to be a card fault but have to wait around a few days to see if the dropout is still there. I'm using a second same make & model card to confirm where the fault is.

 

[delshay]

Sounds like the monitor has bad 'filtering circuitry'. I know on PCIe*, termination resistance and capacitance is very important, and poorly-engineered or damaged components can cause 'issues'.

I think you almost got this right & i'm going to agree with you but not at that location. I think it's at the display output either filtering or display voltage issue, ie voltage could possibly be a little on the low side. This is why a shorter cable works without issues.

I'v been using the old cable since yesterday on second card, same make/model & there has not been a single dropout/black screen in the last 24hr. This clearly says the other card is faulty even thou there is no drop-out with a shorter cable. I can go further than this & put a scopemeter on the HDMI/display port to confirm the difference in output, but a change in cable says it all & I don't think I need to go further other than repair the card. . ...Be aware the faulty card I bought from a miner who claims the card was new never used, & to be fair the card did look brand new..