1.4 Display Port KVM Switch - Dual Monitor - Two Computer
USB Interface (5-gigabit or 10-gigbit):
10-gigabit
Problem:
GPU driver crashes repeatedly when switching inputs resulting a prolonged black screen and system hang. It usually recovers eventually without a reboot. Happens to both input PCs (4080 Super and RTX 6000 Ada).
System log in Event Viewer shows multiple Event 153 on nvlddmkm noting Resetting/Reset/Restart TDR occurred. They always come in that group of 3 errors. Sometimes it will loop dozens of times, sometimes it will happen once and then recover. It doesn’t matter how long it’s been between KVM switch cycles. About 30% chance to occur on any given switch.
Other Notes:
I’ve updated display drivers on both which actually seems to have made the issue significantly worse (previously was around 1 in 20 switch cycles). My 4080 system seems to have a slightly higher incidence rate and usually a longer recovery cycle. Both are Windows 11 systems.
I’ve been having this issue with my RTX 4080 Super. It is connected to three monitors, but I’ll just list their maximum refresh rates, 180Hz, 240Hz, and 144Hz respectively. The issue is worse in my machine because aside from driver crashes, the machine seemingly restarts on its own, suggesting that the machine itself has crashed.
The only workaround that worked for me so far is to set all my monitors’ refresh rates to just 144Hz. And here’s the kicker: if you don’t set this to every possible monitor configuration, these values can go back to the maximum value. By “configuration”, I mean something like “config 1: monitor 1 is off, monitors 2 and 3 are on”; “config 2: monitor 3 is off, monitors 1 and 2 are on”. Well, this is only a concern for those people who have multi-monitor configurations.
@wendell
Thanks for the advice! I was already using very high quality 1m cables all around. I made that change when I first attempted to set this up using random cables (freebie pack-ins with monitors) and couldn’t get 240hz to work.
I had just stopped using the KVM for displays and just manually changed display inputs over the past few months to avoid this issue. I would have just continued with that if both of my monitors supported input selection over DDC/CI, but LG really hates standards. I finally had some free cycles recently to debug this and made some significant progress:
With the latest drivers I still get the issue, but now it also BSODs sometimes. That’s actually a good thing because it gave me a dump to work with. In windbg I found that the crash was happening in on of the nv driver modules and actually had enough symbols to see that it was in code related to initializing HDR. I’m using two monitors, one HDR one SDR (I need both for my job).
I disabled HDR entirely on one system and the issue went away when switching to that computer. It still occurred switching to the other until I disabled HDR on that computer as well. I then swapped out the 4080 for a 9700XT from another system, and enabled HDR for my OLED display on that system and have no issues switching with the KVM in that state.
So I wrote an autohotkey script to enable/disable HDR on my workstation and configured a macro on my Moonlander keyboard to send the “Disable HDR” hotkey before then sending the key sequence to switch the KVM. Then I set the “Enable HDR” command to run a few seconds after switching back and also on login because the hotkey won’t work on the login screen. Certainly not ideal, but I don’t actually have to interact with it each time I switch so it’s a fine workaround. I have not had any hangs or crashes with this setup.
Not sure if this is something you might be able to inform NVIDIA about, but none of my contacts there (through work) work in that space so anything I report to them just ends up in a generic issues queue that probably goes nowhere, especially given how many specific things need to be in place for the repro: DP1/4 KVM to switch inputs with 2 displays; one HDR, on SDR. There might also be more specifics about my setup that affect this like different refresh rates on my displays, 10bit color depth, custom calibrated icc profiles on everything, etc. so it’s unlikely that they would be testing my use case in there QA process.
