KVM Model:
DP 1.4 Dual-monitor 4 computer
USB Interface (5-gigabit or 10-gigbit):
type here
Problem:
-
Switching away and back via KVM causes both monitors to go black for ~5 seconds.
-
When displays return:
-
All windows are stacked in the top-left of monitor 1.
-
Windows are resized smaller (as if desktop bounds temporarily shrank).
-
After ~1 second they expand slightly but remain incorrect.
-
-
No Windows device disconnect sound.
-
Monitors do not renumber.
-
Resolution remains 3840×2160 @ 60Hz.
-
Both displays blank and return simultaneously.
Behavior is identical whether switching away for 1 second or 20+ seconds.
Other Notes:
Hardware
-
KVM: Level1Techs DisplayPort 1.4 Dual-Monitor 4-Computer KVM
-
GPU: NVIDIA RTX 4090
-
Monitors: 2 × Dell U2718Q (4K 3840×2160 @ 60Hz)
-
Cables: StarTech DP14VMM3M (3m VESA-certified DP 1.4)
-
OS: Windows 11
-
Driver: NVIDIA Studio Driver 595.71 (also tested Game Ready 595.71)
Connection path:
RTX 4090 → 3m DP → Level1 KVM → DP → Dell U2718Q
Both monitors are connected to the same RTX 4090.
What I’ve Tried
Windows / OS
-
Disabled:
-
“Remember window locations based on monitor connection”
-
“Minimize windows when monitor is disconnected”
-
Snap Assist entirely
-
-
Disabled MPO (OverlayTestMode = 5)
-
Disabled dynamic topology (DisableDynamicTopology = 1)
-
Disabled taskbar auto-hide
-
Verified identical scaling on both monitors
-
Tried PowerToys FancyZones restore behavior
No change.
NVIDIA
-
Switched from Game Ready to Studio driver
-
Clean install performed
-
Disabled G-Sync (not present)
-
Forced:
-
GPU scaling
-
No scaling
-
RGB 8bpc
-
-
Set power management to Prefer Maximum Performance (tested, then reverted)
-
Confirmed 3840×2160 under PC resolution section
No change.
Observations
-
Full ~5 second blackout suggests full DP link retrain.
-
No device disconnect sound suggests no USB/EDID loss.
-
Appears that during retrain the RTX briefly reports reduced desktop bounds, causing DWM to clamp window positions.
Question for Level1
Is this expected behavior due to HPD drop on switch?
Would inserting a DP 1.4 retimer / HPD buffer between GPU and KVM prevent the full display engine teardown?
Or is there a firmware setting on the KVM that reduces retrain time / preserves HPD state more aggressively?