Spectrum ES07D03 on v108 still has no signal issues when turning it off and on

ryao · April 14, 2023, 9:05pm

I have the ES07D03 connected to an Asus GeForce RTX 3080 Strix on my machine running Linux with KDE Plasma 5.27.4. The nvidia driver is version 530.41.03. The monitor is set to 144Hz via KDE’s display configuration and the Nvidia driver is configured with AllowGSYNCCompatible=On in xorg.conf. The display has VRR enabled in the HUD. I am using an “Amazon Basics High-Speed HDMI Cable (48Gbps, 8K/60Hz ) - 3 Feet, White” cable to connect the display’s HDMI port 1 to the graphics card’s HDMI port 0:

When I press the button to turn the display off and then again to turn it on, I get a no signal issue. If I press Ctrl+Alt+F6 to change to a virtual terminal, the virtual terminal appears without a problem. If I press Ctrl+Alt+F7 to switch back to my graphical desktop, it then appears.

If I physically unplug the monitor via AC power and plug it back into AC power, I get a no signal issue. If I press Ctrl+Alt+F6 to change to a virtual terminal, the virtual terminal appears without a problem. If I press Ctrl+Alt+F7 to switch back to my graphical desktop, the no signal issue persists. If I physically disconnect the HDMI cable and plug it in again, then the desktop is properly displayed.

If I open Display Configuration, set the monitor to 60Hz, press the button to turn it off and then again to turn it on, the desktop reappears as it is supposed to appear. This did not occur on earlier firmware versions as far as I can remember.

If I set the monitor to 100Hz or 120Hz via Display Configuration, the same no signal issue occurs. I have not tested other refresh rates.

I tried disabling VRR and the no signal issue persists, so VRR does not appear to be a factor. Interestingly, while testing this, occasionally, I would have an issue where the screen was split along a line. The right side of the line was shifted up such that there was a gray box at the bottom while the left side had a gray box at the top. The HUD claimed that the monitor had a resolution of 3840 x 2223 @ 144Hz. Changing the refresh rate to 120Hz made the issue go away and changing it back to 144Hz had everything appear as normal.

Also, when doing my tests, I once observed the following message printed to dmesg from the nvidia driver:

[109021.706632] nvidia-modeset: WARNING: GPU:0: HDMI FRL link training failed.

We can see where in nvidia’s driver that message originates here:

github.com

NVIDIA/open-gpu-kernel-modules/blob/4397463e738d2d90aa1164cc5948e723701f7b53/src/nvidia-modeset/src/nvkms-hdmi.c#L2167


      
          do {
              ret = NvHdmi_SetFRLConfig(pDevEvo->hdmiLib.handle,
                                        pDispEvo->displayOwner,
                                        pHeadState->activeRmId,
                                        NV_FALSE /* bFakeLt */,
                                        pFrlConfig);
          } while (ret != NVHDMIPKT_SUCCESS && retries++ < MAX_RETRIES);
          
          
if (ret != NVHDMIPKT_SUCCESS) {
              nvEvoLogDisp(pDispEvo, EVO_LOG_WARN,
                           "HDMI FRL link training failed.");
          
          
    /*
               * Link training failed even after retrying.  Since link training
               * happens after we've already committed to a modeset and failing is
               * not an option, try one last time with the 'bFakeLt' parameter
               * set, which should enable enough of the display hardware to
               * prevent hangs when we attempt to drive the OR with
               * PROTOCOL_HDMI_FRL.
               */
              ret = NvHdmi_SetFRLConfig(pDevEvo->hdmiLib.handle,

I notice that 60Hz only needs FRL2, while 100Hz and 120Hz needs FRL4, and 144Hz needs FRL5. The message that the driver printed to dmesg about a FRL training issue suggests that the monitor cannot always reliably do link training needed for FRL4 or higher. There could be other things going wrong too, but the link training message is very suspicious to me.

That said, there is a definite compatibility issue between the ES07D03 at 4K 144Hz, the Nvidia GeForce RTX 3080 (and likely other Nvidia 30 series graphics cards) and Linux. I have been suffering from this since I got my monitor in 2021, so I really would like to see a firmware fix done. I recall reporting it to Nvidia, but they could not reproduce it and presumably they were not using the Eve Spectrum to test this.