PC Won't Go to Sleep? Power Requests, Driver Holds, Wake Sources & Modern Standby Diagnosis Guide

2026-11-12 | GPU Repair | 25 minutes | HK GPU Fix Editorial Team

PC Won't Go to Sleep? Power Requests, Driver Holds, Wake Sources & Modern Standby Diagnosis Guide

Key Takeaways

A PC that will not go to sleep is usually not broken: some program, service, or driver has filed a power request that blocks sleep, and Windows will not say which one. The command powercfg /requests names the holder, and the diagnosis is to follow that chain to its source.

Step 0: Define the Symptom — "Will Not Sleep" Is One Symptom, in Three Forms

Before any testing, pin down what this guide covers. The symptom here is one thing only: the machine will not enter sleep. The sleep timer expires and nothing happens, or you press Sleep by hand and nothing happens, or the machine appears to sleep for a second and comes straight back, or the display goes dark while the machine itself keeps running — fans turning, lights on, drives active. The starting point is "it will not go to sleep", not "it slept and something went wrong after waking".

Three sub-symptoms have to be kept apart, because they point at different layers and need different tools. Mixing them up is the most common error in sleep diagnosis.

  1. It never sleeps at all. The timer expires, or you press Sleep, and the machine carries on exactly as before, with no reaction.
  2. It sleeps for a second and wakes immediately. You press Sleep, the machine appears to enter sleep, and then it comes straight back to the desktop.
  3. The display sleeps, but the machine does not. The screen goes dark, but the case fans keep turning, the lights stay on, and the system never enters a low-power state.

These three look similar from the chair, and they are not the same fault. The first is a power request blocking entry. The second is, strictly speaking, not a failure to enter sleep at all — it is a successful entry followed immediately by a wake event, and it belongs to the wake-source side of the diagnosis. The third is a split between two different power requests, one for the display and one for the system. This guide explains how to tell them apart, and where each one hands over to its own guide.

Sleep, Hibernate, and Shut Down Are Three Different Power States

The first check costs nothing and prevents a whole class of confusion: which state did you actually ask for? Sleep, hibernate, and shut down look similar from the outside, and their power states are completely different. Sleep keeps the machine in a low-power state with memory still powered, ready to wake quickly. Hibernate writes memory to disk and then powers off. Shut down powers off, and under Fast Startup it is a hybrid shut down that restores the kernel from hibernation rather than starting cleanly.

That distinction matters because a dark screen proves nothing on its own. A dark screen can mean the display timed out while the system stayed awake, or the system slept, or the system hibernated, or the system shut down. The evidence that decides it is physical: are the case fans still turning, has the power light changed to its standby state, is the drive light still active? "The screen went dark" is not the same as "the computer went to sleep", and the rest of this guide depends on keeping the two apart.

A related boundary: if the Sleep option itself is missing from the Start menu, that is a different layer again — the firmware may not provide a supported sleep state, or a group policy may have disabled it. That is not the "will not sleep" of this guide, and it is mentioned here only so it is not mistaken for it. On a company machine, sleep settings may also be locked by IT policy and cannot be changed locally; that too is a different layer, noted here and not expanded.

Going to Sleep Is a Negotiation, Not a Single Action

Windows does not simply switch the machine off when you press Sleep. Before entering sleep, it asks every driver and program: are you finished, can we sleep now? Every program, service, and driver can raise a hand and say "not yet, I am still doing something". That raised hand is a power request. As long as one valid power request is active, the system will not sleep — and it will not tell you which program raised its hand. The screen simply stays as it is.

So "I pressed Sleep and nothing happened" does not mean the button is broken, and it does not mean the setting is wrong. Very often it means a program or driver is holding a power request in the background. The work of this guide is to find out who is holding it, and the decisive tool is the command `powercfg /requests`. That command is the centre of this whole guide, and it is worth understanding properly before running it.

Microsoft's own reference wording, as cited in published guides, is that power requests prevent the computer from automatically turning off the display or entering a low-power sleep state. A program that holds a request indefinitely can block sleep indefinitely, and the program is supposed to clear its request when its work is finished. When it does not — or when its work genuinely never finishes — the machine stays awake.

DISPLAY Requests Versus SYSTEM Requests: The Screen Can Sleep While the Machine Stays Awake

Power requests come in categories, and the two that matter most are DISPLAY and SYSTEM. A DISPLAY request asks the display to stay on. A SYSTEM request asks the system not to sleep. A program can hold one without the other, and that split is exactly the mechanism behind the third sub-symptom.

A program holding only a DISPLAY request keeps the screen lit, but the system can still sleep. A program holding a SYSTEM request lets the display time out and go dark, while the system itself is blocked from sleeping. That is why "the display went dark but the machine did not sleep" is not a display fault at all. It is a SYSTEM request holding the system awake while the display's own timeout runs normally. The first question in the diagnosis is therefore always physical: when the screen went dark, had the case fans stopped, and had the power light changed to its standby state? Different answers point at different layers.

The `powercfg /requests` output lists the categories separately — DISPLAY, SYSTEM, AWAYMODE, EXECUTION, and others — and each entry carries a caller type: [PROCESS], [SERVICE], or [DRIVER], followed by a name. Reading that output properly means reading the category and the caller type together. A SYSTEM entry from a media player is a different fact from a DISPLAY entry from the same player, and the action that follows is different too.

Why the Graphics Driver Can Block Sleep

The graphics driver has a central role in this guide, because it sits in the power-state transition in two ways.

First, the driver itself has to change state. The GPU has to move from its working state (D0) into a low-power state (D3) before the machine can sleep. If the driver stalls during that transition, or refuses to release its hold, the system gives up on entering sleep, or the sleep attempt fails and the machine comes straight back out. In the severe case this becomes the blue screen DRIVER_POWER_STATE_FAILURE — a driver power-state transition failure — which belongs to the blue-screen code guide and is mentioned here only in a single sentence.

Second, the driver can hold a request on behalf of another program. Hardware-accelerated programs — a browser playing video, streaming software, recording software — ask the GPU to do work, and the graphics driver files a power request on their behalf. In that case `powercfg /requests` may show the graphics driver's name, but the real source is the application. The driver is only the messenger, not the culprit. Published guidance states this plainly: the driver listed may only be passing along a request that started in an app. Seeing the driver's name in the output and concluding "the driver is broken" is one of the most common errors in sleep diagnosis. The correct move is to follow the chain back to the source program: close the suspect program, run the command again, and see whether the request disappears.

Third, dual monitors and high refresh rates can hold the GPU clock up. With two monitors connected, the GPU memory clock can stay locked at its highest speed and refuse to drop — a normal design behaviour, already documented elsewhere, and not a fault to be treated as one. But that normal behaviour can make the GPU reluctant to enter its low-power state, which in turn can slow or block sleep entry. The boundary has to be stated clearly: a clock that does not drop is not itself a fault, but if it leaves the machine unable to sleep at all, the investigation proceeds from the power-request side, not by blaming the clock.

Manual Sleep Works, Automatic Sleep Does Not: Two Different Pieces of Evidence

This is the most useful split in the whole guide, and it should be tested early. If you can press Sleep by hand and the machine sleeps, but it never sleeps on its own when the timer expires, the problem is very likely in the automatic timing side: some program keeps resetting the idle timer during the waiting period — background media playback, mouse movement, network activity — or some power request only takes effect during automatic sleep. If pressing Sleep by hand also fails, the problem is in the power-request layer, not the timing layer. Published guides state this split explicitly. The two cases need different next steps, and treating them as the same symptom wastes the whole diagnosis.

Sleeps for a Second and Wakes Immediately: Entered, Then Woken

The second sub-symptom — sleeps for a second and wakes immediately — is strictly not "cannot enter sleep". It is "entered, then immediately woken". That distinction decides the tool. A wake event concerns a system that has already entered a low-power state; a power request is a separate issue that prevents entry in the first place. The two must not be confused, and the commands that answer them are different.

For the immediate-wake case, the tool is `powercfg /lastwake`, which reports what woke the machine last time, together with the wake sources: authorised wake devices, wake timers, network wake, and USB devices. That diagnostic tree belongs to the guide about a machine that wakes itself after sleeping, and this guide takes only one sentence from it — the four wake-source categories — and does not rewrite its tree. The job of this guide is to teach the reader how to tell "cannot enter" from "entered and woke at once", and once that split is made, the second case hands over to the wake-source method.

Modern Standby (S0) Versus Traditional Sleep (S3): Two Ways of Sleeping, Two Diagnoses

Not all sleep is the same sleep, and the difference changes the diagnosis.

Traditional sleep, S3, is the machine genuinely sleeping: fans stop, power draw falls very low, and waking means bringing the whole system back. Modern Standby, S0 Low Power Idle, is different: the system appears to sleep, but it can wake briefly to do maintenance — fetching email, applying updates — and then go back to sleep on its own. Which form a machine uses is not the user's choice. It is decided by the firmware and hardware.

Two commands matter here. `powercfg /a` shows which sleep states the machine supports; if its top line shows Standby (S0 Low Power Idle), the machine uses Modern Standby. `powercfg /sleepstudy` writes a report of Modern Standby activity, covering the past three days by default and up to 28 days, which shows whether the machine's brief wake-ups are maintenance or a genuine failure to sleep.

The key distinction is this: under Modern Standby, "wakes briefly and then goes back to sleep on its own" can be normal design, not a fault. The wake guide has already stated this, and this guide repeats it in a single sentence. But "never sleeps at all, from start to finish" is a fault under either mode. Keep the two apart: a normal maintenance wake-up is not a fault. Equally, a fan that turns intermittently under Modern Standby can be normal maintenance; whether it is normal or not is decided by the sleepstudy report, not by the sight of a turning fan.

If `powercfg /a` shows that no sleep state is supported at all, the problem is in the firmware or hardware layer, not in a Windows setting. That is a stop signal, and it is covered in the diagnostic sequence below.

The Diagnostic Sequence

Work through these steps in order, changing one variable at a time. The decisive command has to be run while the problem is actually happening, because it only shows requests that are active at that moment.

Step 0: Confirm the state. Which did you press — sleep, hibernate, or shut down? Three different power states, and the shut-down path under Fast Startup is a hybrid that restores the kernel from hibernation rather than starting cleanly. When the screen goes dark, check the fans, the power light, and the drive light. A dark screen does not prove the computer has entered sleep.

Step 1: Manual versus automatic comparison. Does pressing Sleep by hand work? If manual sleep works but automatic sleep does not, the problem is in the timing or background-activity layer. If manual sleep also fails, the problem is in the power-request layer.

Step 2: The decisive command. Open Terminal or PowerShell as administrator and run `powercfg /requests`. It must be run while the problem is happening, because it only lists requests that are active at that moment. Running it after the problem has passed and seeing nothing proves nothing. Read each category — DISPLAY, SYSTEM, AWAYMODE, EXECUTION — and note each entry's caller type ([PROCESS], [SERVICE], or [DRIVER]) and name. Administrative rights matter: without them the output can come back empty or misleading.

Step 3: Follow the chain to the source. If the entry is a [PROCESS] — a browser playing video, streaming software — close it and run the command again to see whether the request has gone. If the entry is a [DRIVER] — a graphics driver, an audio driver — consider whether it is passing along a request from some program: close the suspect programs and run the command again before blaming the driver. Note that "Legacy Kernel Caller" is a generic label that does not identify the source; it needs further investigation rather than a conclusion.

Step 4: Display versus system split. Display asleep but machine awake points at a SYSTEM request. The whole machine not sleeping at all means reading DISPLAY and SYSTEM together.

Step 5: Immediate-wake split. If the machine enters sleep and wakes at once, switch tools: run `powercfg /lastwake` and investigate wake sources. Do not keep circling the power-request side for a wake problem.

Step 6: Modern Standby comparison. Run `powercfg /a` to see which sleep states are supported. On a Modern Standby machine, run `powercfg /sleepstudy` to read the past three days of activity and judge whether the wake-ups are maintenance or a genuine failure to sleep.

Step 7: Driver comparison. Only update or roll back a driver that `powercfg /requests` or the energy report has actually named. The energy report is `powercfg /energy`, which runs for 60 seconds and checks power settings and drivers. Do not update every driver in the machine before any of them has been named — that changes many variables at once and makes the result impossible to attribute. Graphics, chipset, network, and audio drivers are the ones most commonly named in sleep problems, but "commonly" is not a diagnosis: wait for the output to name one.

Step 8: The stop signals. If `powercfg /requests` shows None in every category, the energy report shows nothing abnormal, manual and automatic sleep both fail, and `powercfg /a` shows no supported sleep state, the problem is in the firmware or hardware layer. Stop changing settings at random and have the machine inspected. The override command, `powercfg /requestsoverride`, is a last resort only: it tells Windows to ignore a particular request, and it may only be used once a caller has been identified and confirmed to have no real work running. Its syntax has to match the entry's caller type and request category exactly. It is not a first-step fix, and using it as one only hides the cause.

Reading the Pattern

What you observeWhat it points to
Press Sleep, nothing happens at allA power request is holding the system; run `powercfg /requests` while it is happening
`powercfg /requests` shows a SYSTEM entryThat caller is blocking sleep entry
`powercfg /requests` shows only a DISPLAY entryDisplay is held on; system sleep may still be possible
Display dark, fans still turning, lights onSYSTEM request blocking sleep while the display timeout runs normally
Named caller is a [PROCESS] such as a browser playing mediaClose the program and re-run the command; the program is the source
Named caller is a [DRIVER] such as the graphics driverDriver may be passing along an application's request; close suspect programs and re-run before blaming the driver
All categories show None, but only when run after the problem passedCommand only shows requests active at that moment; run it while the problem is happening
Manual sleep works, automatic sleep never happensTiming or background-activity layer, not the power-request layer
Manual sleep fails tooPower-request layer
Sleeps for a second, then wakes to the desktopWake-source problem; switch to `powercfg /lastwake`
Brief wake-ups under Modern Standby, then back to sleep on its ownPossibly normal maintenance; judge from the sleepstudy report, not from one wake-up
`powercfg /a` shows no supported sleep stateFirmware or hardware layer; not a Windows setting
`powercfg /requests` all None, energy report clean, manual and automatic both failFirmware or hardware layer; stop and have the machine inspected

The Energy Report: A Second Opinion

The energy report deserves a paragraph of its own, because it answers a different question from `powercfg /requests`. The requests command shows what is holding sleep right now. The energy report, `powercfg /energy`, runs for 60 seconds and checks power settings and drivers for problems, including drivers that are known to interfere with sleep. Use it when the requests output is empty but the machine still will not sleep — it is the next place to look, not the first. As with the driver step, its value is that it names a specific driver. A named driver can be updated or rolled back with a reason. An unnamed driver cannot.

Boundary Cases

Sleep that is shallow, with the fan still turning, needs care under Modern Standby: intermittent fan activity can be normal maintenance, and the sleepstudy report is what decides. Do not declare "it is not sleeping" from the sight of a turning fan alone.

Laptop lid behaviour is a power setting, not a sleep-entry fault. Closing the lid no longer means sleeping by default; after the setting is changed, sleep has to be pressed actively. A lid that does not sleep the machine can be a wrong lid setting rather than a broken sleep entry, and the two have to be separated.

Hibernation is not sleep. Hibernate writes memory to disk and powers off, and a failure to hibernate is a disk-space or hibernation-file problem, not this guide's subject. Fast Startup shut down is not sleep either: a shut down under Fast Startup restores the kernel from hibernation, so restarting is the genuine clean start, and a problem that survives shut down but not restart points at the Fast Startup layer.

A machine that crashes or shows a black screen while idle is a different fault again — an idle-state crash, not a refusal to sleep. The two share the word "idle" and little else, and this guide does not cover the crash side.

Common Mistakes

What This Guide Does Not Cover

This guide covers one symptom: the machine will not enter sleep, in its three forms — never sleeps, sleeps a second and wakes, or display sleeps while the machine does not. It does not cover black screens, performance drops, or fan problems after waking from sleep — those happen after a successful sleep, in the opposite direction. It does not cover a machine that enters sleep and then wakes itself later — that is the wake-source guide, and this guide uses only its four wake-source categories. It does not cover resolution or refresh rate changing after waking, or window layouts scrambling after waking. It does not cover fans and lights staying on after shut down — that is standby power after shut down, a different state from sleep. It does not cover crashing or black screens while idle or at low load, and it does not cover Fast Startup making "shut down and start again" fail to clear a problem — that is the Fast Startup guide.

When to Stop Testing and Seek Help

Stop testing and have the machine inspected if `powercfg /requests` shows None in every category, the energy report shows nothing abnormal, manual and automatic sleep both fail, and `powercfg /a` shows no supported sleep state. That combination points at the firmware or hardware layer, and no further combination of settings will change it. Write down what you observed before you go: whether manual sleep worked, the full `powercfg /requests` output, the `powercfg /a` result, and the sleepstudy findings. That record is what turns "my computer will not sleep" into a diagnosis a technician can work with. HK GPU Fix is at Shop 1209, Dali Plaza, Shek Mun. Phone 2186 8010, WhatsApp 6558 6806. The shop's published three-tier repair pricing is HK$400–2,800, with the actual figure confirmed by a quote after inspection.

FAQ

Q1: I press Sleep and nothing happens. Is the Sleep button broken?

Probably not. Before entering sleep, Windows asks every program and driver whether it is finished, and any of them can file a power request that blocks sleep. Windows will not tell you which one raised its hand. Run `powercfg /requests` as administrator while the problem is happening to see who is holding the request.

Q2: What is a power request, and how do I use powercfg /requests to read the output?

A power request is a program, service, or driver telling Windows "not yet, I am still doing something". As long as one valid request is active, the system will not enter sleep. The request can be for the display (DISPLAY) or for the system itself (SYSTEM), and the two are counted separately. Open Terminal or PowerShell as administrator and run `powercfg /requests` — it must be run while the problem is happening, because it only lists requests that are active at that moment. The output lists categories (DISPLAY, SYSTEM, AWAYMODE, EXECUTION), and each entry carries a caller type: [PROCESS] for a program, [SERVICE] for a system service, [DRIVER] for a driver. Read each category in turn and note each entry's type and name.

Q3: The requests output names my graphics driver. Does that mean the driver is broken — and will updating every driver make it sleep?

Not necessarily. The driver may only be passing along a request that started in an application, such as a browser playing video or streaming software using hardware acceleration. Close the suspect programs and run `powercfg /requests` again. If the request disappears, the program was the source and the driver was only the messenger. As for updating drivers: updating every driver before any of them has been named changes many variables at once and makes the result impossible to attribute. Only update or roll back a driver that `powercfg /requests` or the energy report has actually named. The energy report is `powercfg /energy`, which runs for 60 seconds and checks power settings and drivers, naming the ones with problems.

Q4: The screen went dark but the fans are still running. Is the PC asleep?

Almost certainly not. The display and the system sleep separately: a DISPLAY request holds the screen on, while a SYSTEM request holds the system awake. If the screen is dark but the fans keep turning and the lights stay on, a SYSTEM request is blocking sleep while the display's own timeout runs normally. Check the fans and the power light, not just the screen.

Q5: Manual Sleep works, but the PC never sleeps on its own. What does that mean?

It means the problem is probably in the automatic timing side, not the power-request layer. Something keeps resetting the idle timer during the waiting period — background media playback, mouse movement, or network activity — or a request only takes effect during automatic sleep. If manual sleep also fails, the problem is in the power-request layer instead.

Q6: My PC sleeps for a second and then wakes straight back to the desktop. Is that the same problem?

No. That is not a failure to enter sleep; it is a successful entry followed immediately by a wake event. The tool changes: use `powercfg /lastwake` to find what woke the machine, and investigate wake sources — authorised wake devices, wake timers, network wake, and USB devices — rather than circling the power-request side.

Q7: What is the difference between Modern Standby (S0) and traditional sleep (S3) — and under Modern Standby, is waking briefly a fault?

Traditional sleep (S3) powers the system genuinely down: fans stop and power draw falls very low. Modern Standby (S0 Low Power Idle) looks asleep but can wake briefly to do maintenance, such as fetching email, and then go back to sleep on its own. Which form a machine uses is decided by firmware and hardware, not by the user. Run `powercfg /a` to see which states are supported. Under Modern Standby, a brief maintenance wake-up followed by going back to sleep on its own can be normal design, not a fault. Judge it from the sleepstudy report — `powercfg /sleepstudy` covers the past three days by default — looking at frequency and timing, not from a single wake-up. Never sleeping at all, from start to finish, is a fault under either mode.

Q8: powercfg /requests shows None in every category. Does that mean there is no problem?

No. The command only shows requests that are active at the moment it runs, so an intermittent problem has to be caught while it is happening. Seeing None once only means there was no request at that moment. Also note that the command needs administrator rights for complete output — without them it can come back empty or misleading.

Q9: Should I use powercfg /requestsoverride to silence the request that is blocking sleep?

Only as a last resort. The override tells Windows to ignore a particular request; it does not resolve the cause behind it. Use it only once the caller has been identified and confirmed to have no real work running, and make sure the syntax matches the entry's caller type and request category exactly. It is not a first-step fix.

Q10: When should I stop trying settings and get the machine inspected?

Stop if `powercfg /requests` shows None in every category, the energy report shows nothing abnormal, manual and automatic sleep both fail, and `powercfg /a` shows no supported sleep state. That combination points at the firmware or hardware layer. Write down whether manual sleep worked, the full requests output, the `powercfg /a` result, and the sleepstudy findings before you go.

Related articles

Related services

If you need Power connector repair, see these service pages: