Application Has Been Blocked From Accessing Graphics Hardware? TDR Recovery Failure, Per-App Block & Driver/App/Hardware Diagnosis Guide

Key Summary
A message saying an application has been blocked from accessing graphics hardware does not, on its own, mean the graphics card is broken. The block targets one application only, and it is the third ending of the same timeout detection and recovery mechanism. Five layers below separate an application problem, a driver problem, and a hardware problem.
The Blocked Message Is the Third Ending of the Same TDR Story
The message arrives without warning. A game, a browser, or a background program is running, and then Windows shows a notification: the application has been blocked from accessing graphics hardware. The program may close, freeze, or keep running without its graphics acceleration. The rest of the desktop usually keeps working. That last detail is the most important one in the whole problem, and it is the detail most people skip past.
To understand the block, it helps to understand the mechanism behind it. Windows watches the graphics card through a system called Timeout Detection and Recovery, usually shortened to TDR. When a graphics operation takes longer than the expected limit — by default, two seconds — the graphics scheduler in dxgkrnl.sys notices the delay. It first tries to pre-empt the work that is stuck, taking the card back from the operation that has overrun. If pre-emption does not work, the scheduler declares the GPU frozen and resets the driver and the card, so that the whole machine does not lock up. The reset is a protective measure. It is designed to recover from a single stuck operation without forcing a restart.
That mechanism has more than one ending, and the ending is what decides which article a user needs. The same underlying event — the GPU hanging, followed by a TDR reset — has two endings that are already well known: recovery and a crash with an error. This article covers the third ending, the one that is usually missing from the other two stories.
The first ending is recovery. The reset succeeds, the driver comes back, the screen flickers once and returns, and Windows shows a message saying the display driver stopped responding and has recovered. Seeing the message itself means the recovery mechanism has done its work. The program keeps running, or it restarts cleanly, and the user carries on.
The second ending is a crash with an error. The reset happens, but the application cannot survive it, so the game closes with a device-removed error, or the system cannot complete the reset and shows a blue screen. The program is gone, or the machine needs a restart, and the user is left with an error code to look up.
The third ending is the block. The reset does not hold, or one application keeps triggering engine timeouts again and again, and Windows changes strategy. Instead of resetting the driver again and again, it blocks that one application's access to the graphics hardware and records the block, so that the offending program cannot keep pushing the system toward a blue screen. The notification names the blocked program, and from that moment the program cannot use the card. Everything else on the desktop carries on as normal.
The three endings share one cause — a GPU that stopped responding in time — but they point at different next steps. A recovery ending points at the recovery mechanism itself. A crash ending points at the application that could not survive the reset. A block ending points at the specific application that kept triggering timeouts, and at the question of why that one program keeps doing so while others do not. Mixing the endings up leads to the wrong fix: treating a block as though the card had failed leads to replacing hardware that is healthy, and treating a block as though it were a simple driver crash leads to reinstalling a driver that was never the problem.
There is one more distinction worth making at the start. TDR is a symptom, not a cause. The reset, the crash, and the block are all things Windows does in response to a GPU that stopped responding. None of them says, on its own, why the GPU stopped responding. The reason can sit in the application, in the driver, in the recovery layer itself, in the load on the card, or in the system settings around it. The five layers below work through those possibilities in order, and the seven steps after them turn the layers into a sequence of tests.
What the Block Actually Does
The block is narrower than it sounds. It does not disable the graphics card. It does not stop the driver. It does not affect the desktop, the taskbar, or other programs. It stops one application — the one named in the notification — from using the graphics hardware. That narrowness is diagnostic evidence in itself. If the card itself had failed, the failure would not be so selective. A card that cannot render would fail everywhere, not in one program while the rest of the desktop renders normally.
The notification usually names the blocked executable. It can be a game, a browser, or a background program such as nvcontainer.exe. The name matters, because the name is the starting point for every test that follows. A block against a game that was just updated points in a different direction from a block against a browser that has been running the same version for months, and a block against a background program points somewhere different again. The first step in any diagnosis is therefore to write down the full notification text and the executable name exactly as shown, together with what was happening at the time and whether it was the first occurrence. Without that record, everything that follows is guesswork.
The block is also recorded. Windows logs the underlying timeout events, and those records are the only objective evidence of what happened. The Display source in Event Viewer records Event ID 4101, with the text "Display driver nvlddmkm stopped responding and has successfully recovered", when a driver reset takes place. Reliability Monitor records LiveKernelEvent 141 for an engine timeout and LiveKernelEvent 117 for a GPU hang that triggered a recovery. Matching those records against the time the block appeared shows whether a TDR event really happened at that moment, and how often it has been happening. A block with a matching trail of 4101 and LiveKernelEvent 141 entries is a confirmed TDR story. A block with no system-level record at all points more toward the application layer, and the diagnosis should not jump to a hardware conclusion.
There is a related identifier that is often misunderstood, and it is worth clearing up early. Microsoft defines VIDEO_TDR_APPLICATION_BLOCKED, bug-check code 0x00000142, as an identifier for live dumps. Microsoft states plainly that this code can never be used for a real bug check, and that it is used to identify live dumps. It is not a blue-screen code, it is not a stop code a user would see on a blue screen, and it cannot be used as evidence of a hardware fault. Anyone who presents 0x142 as a blue-screen code is describing it wrongly. The identifier exists so that engineers can recognise a live dump taken when an application was blocked; it is not a consumer-facing error.
The block is also different from the other kind of block that Windows can apply to graphics hardware. In Device Manager, Code 48 means a driver has been placed on a block list and Windows refuses to load it at all, so the whole card cannot run. The driver can be completely and correctly signed and still be blocked in that way. That is a driver-layer block: the driver never loads, and the card never starts. The block in this article is an application-layer block: the driver loads, the card runs, the desktop renders, and only one program is refused access. The two blocks start from opposite ends. Code 48 means the driver is stopped before it begins. The application block means the driver is running and one program is stopped after it begins. Confusing the two leads to the wrong repair, because a driver-layer block is fixed at the driver level while an application-layer block is diagnosed at the application level first.
The Five Suspicion Layers
A block can be produced from five layers. Working through them in order is faster than guessing, because each layer has its own test and its own evidence. The layers are the application, the driver, the TDR and recovery layer, the load and stability layer, and the system and settings layer.
Layer 1: The Application Layer
The first question is always which application was blocked, and whether it is the only one. The notification names the executable, and that name is the first clue. If only one application is ever blocked, and every other program that uses the card runs normally, the evidence points at the application layer. A different program running the same kind of content — another game, another browser — is the control test. If the control program runs without any problem, the fault is tied to the blocked program, not to the card or the driver.
Applications can trigger a block in several ways. A program can send an unusual instruction that the driver cannot complete in time, and the engine timeout follows. The device-removed error codes HUNG and RESET describe the same kind of underlying event from the application side: the GPU hung, and then the reset followed. A program that was updated just before the blocks started is a strong candidate, because an update can change the rendering path, the shader code, or the way the program talks to the driver. A program that only blocks under one particular mode — one graphics preset, one resolution, one feature switched on — is also pointing at itself rather than at the hardware.
The application layer also includes background programs. A background program that uses the graphics card for acceleration, recording, or monitoring can be the blocked executable even when the user was looking at a different window. When the notification names a background program, the diagnosis starts with that program, not with the game in the foreground.
Layer 2: The Driver Layer
The second layer is the driver. A driver that does not match the current Windows build is a common layer, and the timing gives it away: blocks that start right after a large Windows update point toward a driver that no longer fits the new build. A driver that is damaged or incompletely installed can also produce timeouts, because the driver is the software that translates the application's instructions into work for the card. If that translation is broken, the card can appear to hang even though the hardware is fine.
Windows can also change the driver in the background, through Windows Update, without the user doing anything. A machine that starts blocking applications after a quiet background driver change is showing a driver-layer symptom, and the fix is a driver comparison, not a hardware replacement. The driver version should always be checked against the current official release, never written down from a fixed number, because version numbers change and a number recorded months ago says nothing about what is current now.
Layer 3: The TDR and Recovery Layer
The third layer is the recovery mechanism itself. The three endings — recovery, crash, and block — are all outcomes of the same mechanism, and the difference between them is in how the recovery goes. A recovery that succeeds once is the first ending. A recovery that fails outright can end in a blue screen, usually VIDEO_TDR_FAILURE, and the machine needs a restart. A recovery that succeeds but has to be repeated again and again for the same application ends in the block: Windows stops resetting and starts refusing.
That repetition is the key condition. A single timeout does not produce a block. The block is what happens when an individual application repeatedly triggers engine timeouts. That is why the event record matters so much: a trail of Event ID 4101 entries and LiveKernelEvent 141 and 117 records, clustered around the same application and the same time of day, shows the repetition that the block is responding to. A single isolated 4101 entry with no repetition points at a one-off hang, not at the pattern that produces a block.
The reset timeout setting, TdrDelay, belongs in this layer, and it needs a clear warning. Changing TdrDelay only changes how long Windows waits before deciding the GPU has hung. It does not fix whatever is making the driver stall. It can make a genuine hang last longer before the reset happens, which is the opposite of a repair. Microsoft describes the related settings as being for testing and debugging purposes, and they should not be changed casually. Changing TdrDelay to a larger value is not a fix, and it should never be presented as the first step in a diagnosis. If the setting is ever changed for a test, the registry should be backed up first, and the change only takes effect after a restart.
Layer 4: The Load and Stability Layer
The fourth layer is load and stability: overclocking, power delivery, and temperature. An unstable overclock can make the card fail to complete work in time, which looks exactly like a driver timeout from the outside. A power supply that cannot hold up under a sudden spike in demand can do the same. A card that only blocks applications under heavy load, and never under light load, is pointing at this layer. A card that blocks applications even when idle is pointing away from it.
The distinction between heavy load and idle matters more than any single reading. A temperature reading on its own proves little, and this guide does not publish a universal temperature limit as a standard, because different cards have different designs and different limits. What matters is the pattern: whether the block follows the load, whether it follows an overclock, and whether it disappears when the card is returned to its factory default clocks. If the block disappears at default clocks, the stability layer was the cause. If it carries on at default clocks, the diagnosis moves on.
Layer 5: The System and Settings Layer
The fifth layer is the system and the settings around the card. Windows graphics settings allow an individual application to be assigned to a particular graphics card, and that assignment is a legitimate control test: moving the blocked application to a different GPU, where one is available, shows whether the block follows the application or the card. An application's own hardware acceleration switch is another control. Switching hardware acceleration off and finding the block gone proves only that the acceleration path is involved in the symptom; it does not prove that the card is broken. The same logic applies to overlay and recording software, which sits between the application and the driver and can change the timing of graphics work. System file damage is sometimes raised as a cause, and tools such as SFC and DISM are sometimes suggested. Those tools should be treated as a system-file layer control test, not as a conclusion: running them and finding nothing does not clear the system, and running them and finding something does not prove the system files were the cause of the block.
Three Endings, One Mechanism
The table below puts the three endings side by side, because the ending is the fastest way to tell which problem a user actually has. The underlying event is the same in all three rows. The difference is what Windows does next.
| Ending | What the user sees | What it means | Where to start |
|---|---|---|---|
| Recovery | "Display driver stopped responding and has successfully recovered", screen flickers and returns | The reset succeeded and the driver came back | The recovery mechanism and the single hang behind it |
| Crash with an error | The game closes with a device-removed error, or the system shows a blue screen, usually VIDEO_TDR_FAILURE | The reset happened but the application could not survive it, or the reset itself failed | The application that crashed, and the reset failure behind it |
| Block | "Application has been blocked from accessing graphics hardware", one program named, the rest of the desktop normal | The application repeatedly triggered engine timeouts, so Windows stopped letting it use the card | The named application, then the driver, then the stability layer |
The block row is the only one where the rest of the desktop keeps working while one program is stopped. That isolation is the defining feature of this article's problem, and it is also the first test. If the whole screen goes black, or the whole machine freezes, the problem is not a per-application block. If only one program is stopped and named, it is.
Two comparisons sharpen the picture further. The first is the other-program comparison: another program running the same kind of content, on the same card, at the same time, without any block. If the control program is fine, the fault is tied to the blocked program. The second is the time comparison: whether the blocks cluster after a driver update, after a Windows update, after an application update, or after an overclock change. Each cluster points at a different layer, and a block with no cluster at all — random, across programs, across loads — points away from the application layer and toward the driver or the stability layer.
Seven-Step Diagnosis
- Write down the full notification and the blocked program. Record the notification text word for word, the executable name named in it, what was happening at the time, and whether it was the first occurrence. Without this record, every later step is guesswork, and a second occurrence cannot be compared with the first.
- Test one program against another. Open a different program that uses the card in the same way — another game, another browser — and run the same kind of content. If only the first program is blocked, the evidence points at the application layer. If every program is blocked, the evidence points at the driver or the system layer.
- Match the events against the time of the block. In Event Viewer, look for Display source Event ID 4101 around the time the block appeared, and in Reliability Monitor look for LiveKernelEvent 141 and LiveKernelEvent 117 in the same window. If the events are there, a TDR event really happened, and the diagnosis follows the recovery layer. If there is no system-level record at all, the evidence leans toward the application layer, and the diagnosis should not jump to a hardware conclusion.
- Use the graphics stack reset as a control test. Pressing Win+Ctrl+Shift+B resets the graphics stack immediately. If the picture returns or the program recovers after the key press, the evidence supports a graphics-stack layer diagnosis. If nothing changes, the evidence points at the application layer or at the startup handshake instead. Pressing the keys and seeing the picture come back does not mean the problem is cured; it only supports a driver-layer diagnosis. The key combination is a control test, not a repair.
- Clean-install the driver and compare versions, changing one variable at a time. Start with the driver installer's built-in clean installation option. Only if that does not help should a full driver removal tool be considered, and its steps are a separate procedure that this guide does not rewrite. If the blocks started after a driver update, roll back to the version before the update as a comparison. Change only one variable at a time, and write down the result of each change. A block that disappears after a clean installation points at the driver layer. A block that continues on the old driver points elsewhere.
- Return the card to default clocks and compare load and temperature behaviour. Switch off any overclock, restore the factory default clocks, and watch whether the block follows heavy load or appears even at idle. Monitor the temperature trend while the card is under load, without treating any single reading as a pass or fail limit. If the block disappears at default clocks, the stability layer was the cause. If it carries on unchanged, continue to the next step.
- Prepare for inspection. Bring the notification text, the blocked executable name, the list of 4101 and LiveKernelEvent 141 times, the driver version, and the list of steps already tried with their results. Stop immediately and seek inspection if the block is accompanied by a blue screen, if the card disappears from Device Manager, if the machine shows no picture at startup, or if there is a burning smell.
Six Misconceptions
Being blocked means the card is broken. Wrong. The block targets one application only, and the rest of the desktop keeps working. The card may be completely healthy. Ask which program was blocked first, before suspecting hardware.
The block is the same thing as Code 48. Wrong. Code 48 is a driver-layer block: the driver is placed on a block list and Windows refuses to load it, so the whole card cannot run. The block in this article is an application-layer block: the driver loads, the card runs, and only one program is refused access. The starting points are completely different.
Making TdrDelay larger fixes the problem. Wrong. A larger timeout only makes Windows wait longer before resetting, and it does not fix whatever is making the driver stall. Microsoft describes the related settings as being for testing and debugging purposes, and they should not be changed casually or treated as a routine first step. If the setting is ever changed, the registry must be backed up first, and the change only takes effect after a restart.
Switching TDR off is a fix. Wrong. Turning the recovery mechanism off only hides the symptom. It turns a crash into a freeze, and the underlying stall is still there, with a higher risk over time. The mechanism exists to stop a single stuck operation from locking the whole machine, and disabling it removes that protection.
Reinstalling the game or program is the universal first step. Wrong. Reinstalling takes the longest of all the steps, and it only fixes one path — damaged program files. Run the program comparison and the event comparison first, and only then decide whether a reinstall is the right move.
Other programs working normally means the hardware is fine. Wrong. Different programs use different rendering and decoding paths. One program running normally cannot be treated as proof that the hardware is healthy, any more than one program being blocked can be treated as proof that it is broken. Switching hardware acceleration off and finding the symptom gone proves only that the acceleration path is involved in the symptom; it does not prove that the card is broken. Each program is its own test, and no single program's result settles the hardware question.
What Not to Do: Safety Boundaries
There are three things not to do while diagnosing a block. First, never turn TDR off — TDR is the protective mechanism that stops a single stuck operation from locking the whole machine, and turning it off only turns a crash into a freeze; it will not stop the block. Second, never edit the registry as a general fix — a wrong registry change can affect the whole system, and changing TdrDelay will not fix the reason the driver stalls. Third, never delete driver folders or system files by hand — deleting the wrong file can break the driver or the system, and drivers should be removed or updated through the official uninstaller or update method. The seven steps above are enough to narrow the cause; none of these three is needed.
Running a program as administrator is sometimes suggested as a fix, and it deserves a careful note. It can be written as a control test at the permission layer: if the symptom follows the administrator identity, that result means something. On its own, it is too weak to be called a repair, and it should not be presented as one. The same applies to SFC and DISM. They are control tests at the system-file layer, not conclusions about the cause.
Prevention and Everyday Habits
Prevention is not about checking after every problem appears; it is about keeping enough comparison material on hand in advance. After a driver update or a major Windows update, open a game or program you normally use and check that it still runs normally. If a problem starts after an update, write down the driver version, the Windows build, and which settings changed, so that the next check has a baseline to compare against.
It helps to keep a habit of noting which program runs normally on which version. A program update, a driver update, or a Windows update can each change the graphics path, and with a baseline you can say it was fine before the update and started after it, instead of guessing from feeling.
On settings, do not change several related settings at once. Change one, watch one, so you know which change had an effect. In the Windows graphics settings, a per-app assignment sends a program to a particular card; note which program was assigned to which card, or a later check will not even know what was changed.
Boundaries: What This Guide Does Not Cover
A block notification is a specific shape, and several nearby problems have different shapes. A program that closes silently back to the desktop with no notification is a crash to desktop, not a block. A black screen while gaming where the sound keeps playing and a restart is needed is a different failure again. A screen that turns into a single solid colour and freezes is another shape. A machine that shows a picture at startup and then goes black as Windows loads is a startup-layer black screen. A machine that crashes or goes black at idle or under low load is a power-state layer problem. A browser or office program that goes black, white, or flickers only when hardware acceleration is switched on, and recovers when acceleration is switched off, is an application acceleration-path problem: the program itself is switching acceleration on and off, while the block in this article is Windows refusing a program. A driver that never loads at all, with Code 48 in Device Manager, is a driver-layer block, not an application-layer block.
Once the ending changes, the problem belongs elsewhere. A message saying the driver stopped responding and has recovered is the recovery ending. A game closing with a device-removed error is the crash ending. A blue screen after a failed reset, usually VIDEO_TDR_FAILURE, is the blue-screen form of the crash ending. Each ending has its own diagnosis, and the steps in this guide are written for the block ending only.
The block also has nothing to do with the cable or the monitor. A block is recorded by Windows against a named program, and the rest of the desktop renders normally on the same cable and the same monitor. If the whole picture is affected, the cable and the monitor are worth testing. If only one program is named and blocked, they are not.
When to Send the Card In
If the seven steps point at the application layer — one program blocked, other programs fine, no system-level event trail, and the block follows that program across driver versions — there is nothing to send in. The program is the problem, and no inspection will change what the program sends to the driver. Inspection is worth considering when the evidence contradicts itself: every program is blocked, the event trail shows repeated 4101 and LiveKernelEvent 141 entries across different programs, the block continues at factory default clocks with a clean driver installation, and the driver and the application are both current. Bring the notification text, the blocked executable name, the 4101 and 141 time list, the driver version, and the steps already tried with their results. A model name on its own is not evidence; the list is.
Stop immediately and seek inspection if the block is accompanied by a blue screen, if the card disappears from Device Manager, if the machine shows no picture at startup, or if there is a burning smell. Those are not application-layer symptoms, and continuing to test through them risks the hardware.
For inspection, the workshop is at Room 1209, Dali Plaza, Shek Mun. Phone 2186 8010, WhatsApp 6558 6806. Inspection fees follow the published three-tier pricing, HK$400–2,800. There is no on-site service and no Mong Kok shop.
FAQ
I got "Application has been blocked from accessing graphics hardware" — does that mean my graphics card is broken?
Not necessarily. The block targets one application only, and the rest of the desktop keeps working. The card may be completely healthy. The first question is always which program was blocked, and only after that should hardware be suspected. A card that has genuinely failed does not fail in one program while rendering the rest of the desktop normally.
What is the difference between "has recovered" and "has been blocked"?
They are two endings of the same TDR mechanism. When recovery succeeds, the driver comes back and the picture returns, and that is "has recovered". When recovery cannot hold — when one application keeps triggering engine timeouts again and again — Windows blocks that application's access to the graphics hardware instead, and that is "has been blocked". The underlying event is the same in both cases: the GPU stopped responding in time, and the reset followed. The difference is what Windows does next.
Why is only one application blocked while the others work fine?
Because the block is aimed at an individual application. The notification usually names the blocked executable, and the block applies to that program alone. Opening another program with the same kind of content is the control test: if the other program runs fine, the evidence points at the application layer. If every program is blocked, the evidence points at the driver or the system layer instead.
Is this block the same as Code 48 in Device Manager?
No. Code 48 is a driver-layer block: the driver is placed on a block list and Windows refuses to load it, so the whole card cannot run. The block in this article is an application-layer block: the driver loads, the card runs, and only one program is refused access. The starting points are completely different, and the two problems are diagnosed and repaired in different places.
What should I look for in Event Viewer?
Look for Display source Event ID 4101, with the text "Display driver nvlddmkm stopped responding and has successfully recovered", and in Reliability Monitor look for LiveKernelEvent 141 and LiveKernelEvent 117. Match those records against the time the block appeared. If the events are there, a TDR event really happened. If there is no system-level record at all, the evidence leans toward the application layer, and the diagnosis should not jump to a hardware conclusion.
Does pressing Win+Ctrl+Shift+B fix it?
No. It is a graphics stack reset control test. If the picture returns or the program recovers after the key press, the evidence supports a graphics-stack layer diagnosis. If nothing changes, the evidence points at the application layer or at the startup handshake instead. Pressing the keys and seeing the picture come back does not mean the problem is cured; it only supports a driver-layer diagnosis. The key combination is a control test, not a repair.
Is increasing TdrDelay a fix?
No. A larger timeout only makes Windows wait longer before resetting, and it does not fix whatever is making the driver stall. It can make a genuine hang last longer before the reset happens. Microsoft describes the related settings as being for testing and debugging purposes, and they should not be changed casually. If the setting is ever changed for a test, the registry must be backed up first, and the change only takes effect after a restart.
Is 0x142 a blue-screen code?
No. Microsoft defines VIDEO_TDR_APPLICATION_BLOCKED, bug-check code 0x00000142, as an identifier for live dumps, and states plainly that this code can never be used for a real bug check. It is not a blue-screen code, it is not a stop code a user would see on a blue screen, and it cannot be used as evidence of a hardware fault. It exists so that engineers can recognise a live dump taken when an application was blocked.
It started after a driver update — what should I do?
Start with a clean installation of the driver, using the driver installer's built-in clean installation option, and then try rolling back to the version before the update as a comparison. Change only one variable at a time, and write down the result of each change. Driver version numbers should be checked against the current official release, never written down from a fixed number, because version numbers change. If the block continues after the update and the rollback, the driver version was not the problem.
What information should I bring if I send it in for inspection?
Bring the notification text, the blocked executable name, the 4101 and LiveKernelEvent 141 time list, the driver version, and the steps already tried with their results. A model name on its own is not evidence — the list is. Stop immediately and seek inspection if the block is accompanied by a blue screen, if the card disappears from Device Manager, if the machine shows no picture at startup, or if there is a burning smell. The workshop is at Room 1209, Dali Plaza, Shek Mun. Phone 2186 8010, WhatsApp 6558 6806. Inspection fees follow the published three-tier pricing, HK$400–2,800.