IMDTouchKnowledge BaseShop main store

Troubleshooting

Why a Browser HMI Tab Goes Black for 20 Seconds

HMI and touchscreens Published 6 min read

Short answer

A whole-tab black screen that clears after 20 to 30 seconds is client memory pressure paging the renderer to disk until proven otherwise, not a gateway, session-timeout or remote tag-provider fault. Capture RAM, committed memory, paging and disk activity on the client during the event before changing any setting; with 16 or more tabs open, the client is the layer to measure first.

On this page
  1. Why the whole tab goes black
  2. Ruling out the timeouts for a 20 to 30 second black screen
  3. What to capture during the event
  4. Reading the capture and fixing the right layer
  5. Why the settings already changed did not help
  6. Proving the fix at the real tab count
  7. Questions asked

A Perspective session that paints solid black for 20 to 30 seconds when you click back into its tab, then recovers on its own, is a client-computer symptom. In the field case behind this article it appeared with 16 or more HMI tabs open, on Windows 10 and Windows Server 2022, in Chrome and in Edge, on Ignition 8.1.44, while the gateway carried roughly 60 active sessions and the affected session stayed alive. That combination points at physical memory in the browsing station: with enough tabs, Windows pages the tab's rendering data out to disk, and reading it back takes seconds. Change no timeout and no browser setting yet. Capture RAM, committed memory, hard faults, disk activity and CPU on the client during one event; that capture, not the gateway configuration, decides the fix.

Why the whole tab goes black #

Each open tab adds a renderer process with its own working set on top of the browser framework, all competing for the station's physical RAM. Once demand passes what the machine holds, Windows moves the least-recently-used pages to disk. The tab stays open and the session stays connected, so nothing looks wrong anywhere but the screen; the moment the tab comes to the front, every page Windows moved must be read back before the renderer can draw a frame. The black interval you time is that read-back and re-render on a disk slow enough to notice.

Tab sleeping and tab discarding are browser policies, and both were already disabled on site. Operating-system paging is not a browser policy: it happens when committed memory across every running process outruns physical RAM, and no browser preference switches it off.

On a dedicated operator panel the same blank-then-recover symptom points into the display hardware; HMI screen goes blank for seconds then recovers covers that case. In a browser tab it points at the computer running the browser.

Ruling out the timeouts for a 20 to 30 second black screen #

Inputs, each with its source:

  • Black-screen duration: 20 to 30 s, reported on site with the tab counts above.
  • Project idle timeout: 720 min, action Close Session, from the project configuration.
  • Session timeout: 5 min, from the same configuration.
  • Perspective client default for the maximum idle time to detect a connection to a Gateway: 120000 ms, from the documented Perspective client parameters.

The rule in words: a timer can only act after its full duration has elapsed, so a timeout explains the black screen only if its duration fits inside the symptom window.

The arithmetic:

  • 720 min × 60 s/min = 43,200 s
  • 5 min × 60 s/min = 300 s
  • 120,000 ms ÷ 1,000 ms/s = 120 s
  • Longest observed window 30 s against the smallest timer 120 s: 120 ÷ 30 = 4

The result: the shortest timer anywhere in the chain outlasts the longest black screen by a factor of four, so nothing can expire inside the window.

The decision: a black screen shorter than the smallest active timer is not a timeout event, and here the boundary sits at 120 s while the symptom stops at 30 s. Leave the 720-minute idle timeout and the 5-minute session timeout exactly as they are; neither raising nor lowering them can remove a delay they are too long to cause. The roughly 60 sessions on the gateway describe load, not this symptom, because the failure follows the tab you switch to, not the session.

What to capture during the event #

Reproduce this on a bench client with test tags, never on a station an operator needs while the machine runs. Log against one clock and note the exact timestamp of the switch back into the tab:

  • Client memory: physical RAM in use, committed memory against its limit, and hard faults per second, sampled every second through the event.
  • Client disk and CPU: disk queue length and activity, plus CPU, to separate paging from a busy processor.
  • Browser detail: memory and CPU per renderer in the browser's own task manager, to tell one constrained tab from a constrained process tree.
  • Gateway state at the same timestamp: whether the session is still active, and whether any other client shows anything at that moment.
  • Data path: whether displayed values are stale when the page returns, and whether other pages using the same remote provider misbehave.
  • Reproduction variables: tab count, browser, operating system, and whether the station was busy elsewhere just before the switch.

Run once at the operator's real tab count, then at a reduced count, changing one variable at a time. The hardware-acceleration change already made in Chrome and Edge belongs in that sequence too, because no result from it has been recorded.

Reading the capture and fixing the right layer #

What changes during the black screen Layer at fault Fix
Hard faults and disk queue climb on one station; the session stays active in the gateway Client memory pressure Cut the concurrent tab count or fit more RAM in the station, then retest
The session closes in the gateway at that timestamp Session lifecycle Correct the timeout and logout configuration
Several clients stall at the same moment Gateway or network Check gateway load and the client-to-gateway path
Pages render but values freeze across clients Remote tag provider route Check provider and proxy health

When the station doing the paging is also the machine's operator station, replace it with hardware from industrial panel PCs sized for the real tab count rather than for a single full-screen HMI. When values freeze while pages still render, the provider route is the layer, and remote tag history stopping when gateway redundancy fails works that fault in detail.

Why the settings already changed did not help #

Memory Saver sits under More, then Settings, then Performance in Chrome, and switching it off stops Chrome from sleeping tabs itself; the equivalent sleeping-tabs control in Edge was checked on site. Neither touches Windows paging, which proceeds regardless of browser preference once committed memory outruns physical RAM. Hardware acceleration selects the rendering path, not the amount of memory a tab needs, so it does not address paging either; with no recorded result from that change, it stays a test variable. The same memory mechanism at its most visible appears as a virtual-memory warning on Windows HMI stations, and fixing a panel PC virtual memory warning in web HMIs walks through that case.

Proving the fix at the real tab count #

Repeat the exact workload that produced the symptom, same tab count, same pages, same switch pattern, and compare against the baseline capture. The fix counts only when the black interval disappears, the counters that climbed during it stay flat, and the improvement tracks the single layer you changed. A return to the tab with no hard-fault burst, after lowering the tab count or adding RAM, closes the case at the client. A black screen that survives unchanged on clean client counters sends you to the gateway and provider observations on the same timeline.

Questions asked

Could the 720-minute idle timeout or the 5-minute session timeout cause the black screen?

No. Converted to seconds they are 43,200 and 300, and a timer must run its full duration before it does anything. The longest reported black screen, 30 seconds, is one tenth of the smaller value, so neither can fire inside the window. The project inactivity timeout is a separate setting, disabled by default at 0 and set under Project Properties, Perspective, Inactivity, with 900 seconds a common choice; it is just as unable to act in 30 seconds.

Why does the tab still go black with Chrome's Memory Saver switched off?

Memory Saver, under More, Settings, Performance in Chrome, governs browser-initiated sleeping and discarding of tabs. Windows paging is a different mechanism: when committed memory across every process passes what the physical RAM holds, the operating system writes pages to disk whatever the browser prefers, and reading them back on your return is the delay you time. Switching sleeping off removes one cause; it cannot remove that one.

Was turning hardware acceleration off in Chrome and Edge the right move?

It changes the rendering path, not how much memory each tab needs, so it does not address paging. No improvement was recorded from that change in the field case behind this article, which makes it a test variable rather than a fix. Keep it in the one-variable-at-a-time sequence: if repeated runs with it off show the same black interval and the same paging counters, the rendering path is exonerated and the memory layer stands.

Which observations settle that the client is the layer at fault?

Sustained hard faults per second and rising disk queue time on that one station, sampled while the gateway still lists the session as active, is the pairing that identifies memory pressure. One without the other means something else: a closed session points at lifecycle configuration, flat counters with a shared stall points at the gateway or network. The browser's own task manager then tells you whether a single renderer or the whole tree is constrained.

How many HMI tabs can an operator station hold before this starts?

There is no documented per-tab limit for a Perspective client; the ceiling comes from the station's physical RAM and whatever else runs on it. Find it empirically: run the real tab count and watch committed memory against its limit. If committed memory sits at the limit before the switch, the station cannot hold that workload and either the tab count or the RAM has to change.

Keep reading

  • Troubleshooting

    How to Recover a WinCC On-Screen Keyboard Placed Off-Screen

    Set PosLeft, PosTop, PosRight and PosBottom to 0 under the Windows account that runs the HMI, then restart Windows: the keyboard reopens at a sensible default. The values live under HKEY_CURRENT_USER\SOFTWARE\Siemens\SimaticHmiIM on a documented WinCC V17 station, but the subkey varies, so search for PosLeft under HKCU\SOFTWARE\Siemens when the key is not there.

    HMI and touchscreens

  • Explainer

    Where JavaScript and WebAssembly Actually Run on an HMI

    JavaScript and WebAssembly execute in the HMI client's browser engine — the panel itself, a panel PC, or a desktop browser — never inside the PLC, which only serves the files and exposes data through a documented interface. Decide on the rendering route the installed runtime provides, then verify rendering, module execution, and the data path in that order.

    HMI and touchscreens

  • How-to

    Run an HMI application touch-only in forced full screen

    Enable the built-in on-screen keyboard with System.OnScreenKeyboard = 3 so page and popup-dialog entry both work by finger, and launch the runtime through a shortcut whose Target field ends in the -f switch to force full screen. Because -f cannot be exited in-session, keep a second shortcut without it for development, and pass every page and entry field by touch before removing the keyboard and mouse.

    HMI and touchscreens