HDMI/DisplayPort Picture but No Sound? GPU Audio Controller, Audio Endpoint Switching & Speaker Routing Diagnosis Guide

Key Takeaways
When sound keeps moving between devices, the fault is usually not a broken device. It is a default handover that did not move back after an event: the display slept, reconnected, or changed input, a headset was plugged in or removed, or a call started or ended, and the operating system left playback on the substitute device. Diagnose by event, not by device: record the defaults before and after the event, reproduce the event, and only then change one thing at a time.
Opening: Define the Phenomenon First
This guide deals with one specific pattern, and the first step is to confirm that the pattern is really what is happening. Sound worked, and then it moved. At one moment it came from the expected speakers. After some event, it came from somewhere else, or from nowhere the listener expected, while the picture stayed perfect throughout. The move is tied to time: it follows a sleep and wake, a reconnection, an input switch, a headset being plugged in or pulled out, or a call starting or ending. Between events, sound is stable. That time link is the whole diagnosis. If there is no time link, this guide does not apply.
There is one exclusion that must be stated before anything else, because it decides which guide the reader needs. If sound never worked through this connection at all, from the very first attempt, that is a static no-sound chain, and it is covered in full by article #170, which covers the static chain in full. Do not use this guide to re-diagnose a connection that never sounded. This guide starts where #170 ends: sound did work, and then it moved after an event.
A second exclusion matters just as much. If the picture itself is unstable, flashing, dropping out, or missing, stop here and treat the picture first. The picture faults belong to other guides: no signal, intermittent black screens, and display identification problems. In this guide the picture is assumed stable throughout, and picture symptoms are mentioned only to route the reader away. The subject is the audio endpoint moving, not the video signal failing.
The most common misreading of this pattern is to replace hardware. A monitor that sounded yesterday and is silent today is usually not broken, and a headset that "stole" the sound is usually not faulty either. Both devices are behaving as the system instructed them to. The system moved the default, or a program kept the routing it was given during an event, and nobody moved it back. Replacing the monitor, the cable, or the headset does not change that instruction, so the same move happens again with the new hardware. The fix is to find the event and the handover rule behind it, then set a routing that survives the event or a habit that restores it in one step.
Core Concept: Why the Default Does Not Move Back
To understand the move, it helps to separate three ideas that are often blurred together: the audio endpoint, the default playback device, and the default communication device. An endpoint is a named destination the system can send sound to, such as a monitor, a television, a headset, or the motherboard audio output. The default playback device is the endpoint that ordinary programs use when they do not choose one themselves. The default communication device is a separate choice used by calls and voice programs. These are three layers, and an event can move one layer without moving the others, which is why sound can vanish from the speakers while a call still works, or a call can stay on a headset while music returns to the speakers.
When a display sleeps deeply, disconnects, or changes input, its audio endpoint can disappear from the system for a short time. The system cannot send sound to an endpoint that is not there, so it hands playback to a substitute endpoint that is present. That handover is a protective behaviour, not a fault: without it, sound would be lost entirely every time a display slept. The problem is the return journey. When the display wakes or reconnects and its endpoint appears again, the system does not guarantee that playback moves back to it. The substitute may remain the default until someone changes it by hand. From the user's side, this looks like sound "moved" or "broke" after waking the display. From the system's side, it simply never received an instruction to move back.
The same asymmetry explains why the pattern repeats. Every sleep, reconnection, input switch, or device change is another chance for a handover in one direction without a matching handover back. Over a day, the default can drift from device to device, each drift caused by a different event, until the listener no longer knows which device is currently default. That is why a single screenshot of the current default is almost useless as evidence. What matters is the pair of observations: what the default was before the event, and what it became after the event. Without that pair, every later step is guesswork.
A further complication is that individual programs can hold their own routing. Some programs remember the endpoint they were using when they started, and keep using it even after the system default changes. Others follow the system default live. After an event, therefore, two programs can behave differently on the same machine: one still sounds from the old device, the other has already moved to the substitute. That split is not a contradiction and not a sign that one program is broken. It is the expected result of two routing rules operating at once, and the diagnosis must check programs separately rather than assuming one test covers all sound.
Mechanism One: Sleep, Wake, and Reconnection Handover
The most frequent trigger is also the easiest to misread, because the display looks unchanged after it wakes. During deep sleep, the display can stop presenting itself to the system. The system treats that as a disappearance, even though nothing was unplugged. Playback moves to a substitute. When the display wakes, the picture returns, the desktop returns, and the user reasonably assumes everything returned, including the audio routing. But audio routing is not part of the picture returning. The endpoint reappears, yet the default may still sit on the substitute, and the listener hears nothing from the expected speakers while the picture looks normal.
Reconnection behaves the same way, with one extra wrinkle. Unplugging and replugging a display, switching a display input, or switching a switcher or dock input are all, from the system's point of view, a disappearance followed by a reappearance. The handover on disappearance is automatic. The return on reappearance is not guaranteed. That is why the same listener can report that sound "survives" a short sleep but "breaks" after a longer sleep or an input switch: the difference is not the display, it is whether the disappearance lasted long enough, or was complete enough, for the system to move the default. Short events may pass without a handover; long or complete events trigger one.
The diagnostic consequence is strict. Never test audio routing only while everything is awake and stable, because that is precisely the state in which the fault is invisible. The test must include the event itself: record the defaults, trigger the sleep or reconnection, wait for the picture to return, and then read the defaults again. If the defaults changed across the event, the event is confirmed as the trigger. If they did not change, the event is excluded, and the next candidate event is tested the same way. A diagnosis that skips the event step can only describe the current state, never the cause.
Mechanism Two: Many Endpoints, Always a Substitute
A modern machine rarely has one audio endpoint. A monitor, a television, a headset, a USB audio device, and the motherboard audio output can all be present at once, each offering itself as a destination. That abundance is what makes the handover possible: there is always a substitute ready to receive playback when the expected endpoint disappears. It is also what makes the fault confusing, because the substitute is often a device the listener forgot was connected, such as a monitor in another room, a television input, or a headset left plugged in.
The substitute is not chosen at random, but its choice is not under the listener's control either. The system picks among endpoints that are present and ready, and the listener's expectation plays no part in that choice. This is why the same event can send sound to different substitutes on different days: the set of present endpoints changed, because a headset was left plugged in, a television was switched on, or a USB device was connected. The diagnosis must therefore record the full set of present endpoints at the time of the event, not just the expected one and the one that actually sounded. An endpoint that was present but silent can be the next substitute, and it will matter on the next event.
A related trap is to treat the substitute as the fault. Turning the substitute off, unplugging it, or disabling it can make the current event pass, but it does not fix the handover rule. On the next event, with a different set of present endpoints, playback will move to yet another substitute. The durable question is not "how do I silence the substitute" but "what rule moved playback here, and what restores it after this kind of event". That question is answered by the event test, not by removing devices one by one.
Mechanism Three: Monitor Speaker and Headphone-Jack Routing
Monitors complicate routing because a single monitor can offer more than one way to hear sound. Some monitors have built-in speakers. Others have no speakers at all, and instead pass the sound they receive onward through a headphone jack to external speakers or earphones. From the system's side, both kinds present an audio endpoint with the monitor's name, so the two cases look identical in a device list. From the listener's side, they behave very differently, and the difference changes with events.
When a monitor has built-in speakers, sound from its endpoint comes from the monitor itself, subject to the monitor's own volume and mute state, which are separate from the computer's volume. When a monitor has only a headphone jack, sound from its endpoint goes nowhere audible until something is plugged into that jack. A listener who expects the monitor to sound, but has nothing plugged into its jack, will hear silence even though the system routing is correct. The event dimension is that plugging something into the monitor's jack, or unplugging it, can reroute sound within the monitor's own chain, while the system's endpoint name stays the same. The system list shows no change, yet the audible result changes completely.
The diagnostic rule for this mechanism is to test the monitor's own chain separately from the system's chain. First, confirm which kind of monitor it is: built-in speakers, or jack-only onward routing. Then test the jack path on its own: with external speakers or earphones plugged into the monitor's jack, does sound arrive there when the monitor endpoint is default? If yes, the system routing is correct and the question is purely which physical output the monitor is using. If no, the question returns to the system routing and the event log. Conflating these two chains, testing only the system list or only the monitor's speakers, is the most common reason this mechanism is misdiagnosed. The two chains must be tested in order, system first or monitor first, but never merged into a single guess.
Another event-driven wrinkle belongs here. Some monitors reset their own audio path when they change input or wake from sleep, so that sound which previously came from their speakers now waits at their jack, or vice versa. The system endpoint name does not change, so the device list looks innocent. Only the event log, paired with a test of both physical outputs after the event, reveals the change. When a listener reports that sound "moved inside the monitor" after waking it, this is the mechanism to check before touching any system setting.
Mechanism Four: Plugging In and Pulling Out Moves the Default Immediately
Device insertion and removal are the most visible triggers, and the most likely to be misread as faults in the device itself. Plugging in a headset, a USB audio device, or an external speaker set can move the default to the new device immediately, because the system treats a newly arrived endpoint as the listener's current intent. Pulling a device out moves the default again, to whatever remains. Both moves are immediate, and both are one-directional: plugging a headset in moves playback to it, but pulling the headset out does not guarantee playback returns to the previous device. It returns to a present endpoint, which may or may not be the one the listener expected.
The practical consequence is that the order of plugging matters as much as the devices themselves. A listener who plugs a headset in for a call, then pulls it out, then wakes a display, has triggered three handovers in sequence, and the final default is the product of all three, not of any single one. Testing only the final state, without the sequence, makes the result look random. The event log must therefore record insertions and removals in order, with the default read after each one. A log that records only "headset was used today" cannot distinguish a headset handover from a display handover, and the two have different remedies.
There is also a subtler variant. Some devices present more than one endpoint, or present and withdraw endpoints as they change state, for example when a wireless headset connects, disconnects, or switches mode. Each presentation or withdrawal is an event of the same kind as plugging in or pulling out, even though nothing was physically touched. Listeners who report sound moving "by itself" while a wireless device is nearby should check this mechanism first, by recording which endpoints appear and disappear around the move, before suspecting the display or the system.
Mechanism Five: Call Defaults During and After Calls
Calls add a separate layer, because the default communication device can move independently of the default playback device. When a call starts, voice programs use the communication device. When the call ends, ordinary playback should return to the playback device, but the two layers do not always resynchronise cleanly. A listener can finish a call and find that music or video sound now comes from the headset used for the call, even though the call itself is over. Or the reverse: the call stays on the headset correctly, but the listener, testing with music, concludes that "all sound" moved, when only the playback layer moved and the communication layer is behaving correctly.
The diagnostic discipline here is to test the two layers separately, with two different sound sources: a call or voice test for the communication layer, and music or video for the playback layer, each checked before and after a call. A single test, using only one source, cannot separate the layers, and will produce a misleading conclusion in at least half the cases. The event log must also record the call itself as an event, with its start and end marked, because a move that happens during a call and a move that happens at a call's end point to different handover rules.
A common misconception deserves explicit correction. The fact that a headset "takes over" during a call is not a fault in the headset, the voice program, or the system. It is the communication layer doing its job. The fault, if there is one, is only in the return after the call ends, and only if the return fails repeatedly for the same kind of call. A single failure after a single call, with no repeat on a controlled retest, is not yet a diagnosed cause. It is a candidate to be retested, with the layers separated, before any setting is changed permanently.
Mechanism Six: Exclusive Holds Change Across Events
Some programs take an exclusive hold on an endpoint while they play, so that other programs cannot use it at the same time. An event that removes the endpoint, even briefly, can break that hold. When the endpoint returns, the program may resume on a different endpoint, or may keep silent until restarted, while other programs that never held the endpoint move normally. The result is a split that looks like a fault in the program, when the cause is the event breaking its hold.
The test for this mechanism is deliberately narrow. It is confirmed only when the same program behaves differently depending on whether it was playing across the event or started after the event. If a program started after the event sounds correctly from the expected device, while the same program left playing across the event does not, the hold is implicated. If both cases fail identically, the hold is not the explanation, and the diagnosis returns to the default handover. This distinction matters because the remedy differs: a hold problem is managed by restarting the program after the event or by changing how it holds the device, while a default problem is managed by restoring the default. Treating one as the other wastes the listener's time and leaves the real cause in place.
Diagnosis Sequence: Event First, One Variable at a Time
The sequence below is ordered, and its order is part of the method. Each step changes one variable only, and the original setting is recorded before any change, so that every change can be undone. A step is only counted as confirming a cause when the problem disappears after the change and returns after the setting is restored. A problem that disappears after several changes made at once has not been diagnosed; it has only been hidden.
| Step | What to record or change | What it proves |
|---|---|---|
| 1. Baseline | Expected device, current playback default, current communication default, full list of present endpoints, and which physical output the monitor uses | Without this, no later change can be interpreted |
| 2. Event log | Every sleep, wake, reconnection, input switch, plug-in, plug-out, and call start and end, in order, with defaults read after each | Whether the move follows an event, and which kind |
| 3. Reproduce the event | Trigger the same kind of event under control, then read both defaults again | Whether this event kind causes the move, or is excluded |
| 4. Monitor chain test | Test the monitor's speakers and its headphone-jack path separately after the event | Whether the change is inside the monitor or in the system |
| 5. Per-program test | Test playback sound and call sound separately after the event | Whether the playback layer, the communication layer, or both moved |
| 6. Hold test | Compare a program left playing across the event with the same program started after it | Whether an exclusive hold was broken by the event |
| 7. Restore test | Restore the expected default by hand, repeat the event, and read the defaults again | Whether the handover repeats, which is what makes it a cause |
| 8. Boundary check | If sound never worked on this connection, or the picture is unstable, stop and route to #170 or the picture guides | Whether this guide is even the right guide |
Two rules govern the whole sequence. First, never change a setting before the event log is complete, because an early change destroys the evidence the log needs. Second, never declare a cause from a single disappearance. A cause is a repeatable handover: the same event kind, the same move, on a controlled repeat. Everything else is a candidate, and candidates are reported as candidates, not as findings.
Common Misconceptions
"The monitor is broken because it sounded yesterday and is silent today." A monitor whose picture is perfect and whose endpoint is present is rarely the cause of a move that follows an event. Check the event log and the monitor's own speaker-versus-jack path before suspecting the monitor.
"The headset stole my sound." A headset that becomes default when plugged in is following the system's handover rule. The question is not the headset, but whether the default returns after the headset is removed, and after the next event.
"One test with one program proves the routing." Playback routing and communication routing are separate layers, and programs can hold their own routing. Test both layers, and test programs started before and after the event separately.
"Disabling the substitute will fix it." Removing one substitute only changes which substitute receives the next handover. The handover rule remains, and the next event will move sound to another present endpoint.
"The system should always move sound back automatically." That expectation is understandable, but it is not the system's guarantee. The return journey is not automatic in the way the outward handover is, and planning around that asymmetry, by recording defaults and restoring them in one step, is more reliable than assuming a return that may not come.
"If the device list looks right, the routing is right." A device list taken while everything is awake cannot show a handover that happened during sleep or reconnection, and it cannot show a change inside the monitor's own speaker-versus-jack path. Only the before-and-after pair around an event shows those changes.
"Replacing the cable or the device will stop the moving." Replacement changes the hardware, not the handover rule. If the event log shows the move following events with the old hardware, the same events will move sound with new hardware too, until the rule is addressed.
Boundary: What Does Not Belong in This Guide
The following cases are deliberately excluded, and each is routed rather than expanded, so that this guide stays about moving endpoints and does not grow into a general audio guide. A connection on which sound never worked from the start belongs to #170, which covers the static chain in full. A picture that is missing, flashing, or dropping out belongs to the picture guides, including no-signal and intermittent black screen guides. A virtual-reality headset whose picture and tracking are also affected belongs to the virtual-reality guide, not to this one. A display that is misidentified so that its modes are missing belongs to the display identification guide. In each case, only the routing question that overlaps with this guide is mentioned, and the rest is left to the guide that owns it.
When Inspection Is the Right Step
Most moving-endpoint cases are resolved by the event log and the restore test, without any repair. Inspection becomes the right step when the same event kind repeatedly moves sound even after the defaults are restored correctly, when an endpoint disappears and never returns without a restart, or when the monitor's own speaker-versus-jack path behaves inconsistently after the same event. In those cases, bring the event log itself: the list of events in order, the defaults before and after each event, which programs were tested on each layer, and whether the restore test repeated the move. That log is worth more than a description of the current state, because the fault only exists across events, and the log is the only record of it.
HK GPU Fix's published three-tier charge range is HK$400–2,800, subject to a written quote after inspection. The shop is at Room 1209, Dali Plaza, Shek Mun, telephone 2186 8010, WhatsApp 6558 6806. There is no Mong Kok shop and no on-site service.
FAQ
Why does sound move after my display wakes, but not after a short sleep?
A short event may pass without the system moving the default, while a longer or more complete disappearance triggers the handover. The difference is whether the endpoint disappeared long enough for the system to substitute another device. Only the before-and-after defaults around each event kind can tell which kind triggers the move on this machine.
Why does sound go to a device I forgot was connected?
Because the system chooses among endpoints that are present, and a forgotten monitor, television, or headset is still present. The event log should record the full set of present endpoints at the time of the event, not just the expected device.
Why does one program keep sounding from the old device while another moved?
Programs can hold their own routing, and an event can break an exclusive hold for one program while others follow the new default. Test programs separately, and compare a program left playing across the event with the same program started after it.
Why does sound stay on my headset after a call ends?
The communication layer and the playback layer are separate. A call can leave the playback layer on the headset even though the call itself is over. Test the two layers with different sound sources, before and after a call, before changing any setting.
Why does my monitor show up as an audio device but make no sound?
Because a monitor endpoint does not guarantee built-in speakers. Some monitors only pass sound onward through a headphone jack. Test the monitor's speakers and its jack path separately: if sound arrives at the jack, the system routing is correct and the question is which physical output the monitor is using.
If sound never worked on this connection, should I follow this guide?
No. A connection that never sounded is a static chain problem, covered in full by article #170. This guide is only for sound that worked and then moved after an event. Starting with the wrong guide wastes the event log on a fault that has no event in it.
What should I record before asking for inspection?
The events in order, the playback and communication defaults before and after each event, the full list of present endpoints, the results of the monitor speaker-versus-jack test, and whether the restore test repeated the move. Bring that log, because the fault only appears across events.