GPU-Z / HWiNFO Sensor Readings All Blank or Zero? Reporting Path, Tool Permissions & Sensor-Layer Diagnosis Guide

2026-11-13 | GPU Repair | 44 minutes | HK GPU Fix Editorial Team

GPU-Z / HWiNFO Sensor Readings All Blank or Zero? Reporting Path, Tool Permissions & Sensor-Layer Diagnosis Guide

Key Takeaway

GPU-Z or HWiNFO showing every sensor blank, zero, or N/A while the card name reads correctly and games run normally points at the reporting path, not the card. This guide explains the five-layer reporting chain and walks through seven diagnostic steps to find where the readings break — before concluding anything about the hardware.

What This Problem Is — and What It Is Not

Open GPU-Z, click the Sensors tab, and every field is blank, or zero, or marked N/A. Temperature: blank. Core clock: 0 MHz. Memory clock: 0 MHz. Power draw: 0 W. Fan speed: 0 RPM, or N/A. GPU load: 0%. Open HWiNFO and the sensor panel is the same story: every sensor shows N/A, and the values never update. And yet the card's name reads correctly in GPU-Z. The desktop displays normally. Games run. Nothing is visibly wrong with the machine except that the instruments have gone dark.

That scene is the problem this guide covers, and the shape of it is the most important piece of evidence in the whole diagnosis. Every channel is missing at once. Not one temperature reading that looks wrong. Not one fan speed that disagrees with reality. All of them, together, blank or zero. When a single channel is wrong, the fault is usually in that channel. When every channel is missing at once, the fault is usually in something all the channels share — the path the readings travel along, from the card to the screen. This guide is about that path, and about how to find which part of it broke.

Four conditions must all be true before the steps below apply:

  1. The card's identity reads correctly: GPU-Z shows the card's real name and model, not "Unknown", and Device Manager shows the card under its full model name. If the name itself cannot be read, that is a different problem — the identity guide on this site — and the steps below do not apply.
  2. Temperature, clock speed, power draw, fan speed, and load are all missing together — blank, zero, or N/A across the whole panel — not just one of them. A single missing or wrong channel belongs to its own guide: a single temperature channel to the temperature-sensor guide, a single fan-speed channel to the fan-speed guide, a single load reading to the utilisation guide.
  3. The display works and games run. The card is doing its job; only the reporting of what it is doing has stopped. If there is no picture, or games will not start, the problem is not a reporting problem, and a different guide applies.
  4. The reader is looking at a reporting problem, not a hardware problem that happens to share a symptom. A whole panel of missing readings is, on its own, never evidence that the card is faulty. That is the first misconception, and it is dealt with in full below.

The Decisive Fork: The Whole Panel at Once, or One Channel Alone

Before any step is taken, the reader must answer one question, because the answer decides which guide applies. Are all the sensor channels missing together, or is only one channel wrong?

The whole panel at once — temperature, clocks, power, fan speed, and load all blank, zero, or N/A together — is this guide. The channels share a reporting path, and when the path breaks, everything that travels on it goes dark at the same time. The shape of the failure is the diagnosis: a shared path has failed.

One channel alone is a different guide, and the distinction matters because the causes are different. A single temperature channel reading -40°C, or stuck at one value, or jumping wildly, belongs to the temperature-sensor guide: that is one sensing point misbehaving, and the four-level reading chain — the on-die diode, the firmware conversion, the driver interface, and the display software — is the framework for it. A single fan-speed channel reading 0 RPM while the fan is visibly turning belongs to the fan-speed guide: the drive path and the reporting path are two separate paths, and the speed-reporting wire can fail while the fan keeps turning. A single utilisation reading of 0% while a game runs belongs to the utilisation guide: the monitor may be showing the busiest engine rather than the average, or the wrong card may have been selected. Each of those is a single channel with its own cause. None of them is this guide.

There is a related shape that also does not belong here: readings that exist but disagree between two programs. If GPU-Z says one temperature and HWiNFO says another, both programs have readings — they simply disagree, because they sample at different points, poll at different intervals, or define power differently. That disagreement belongs to the guide about why monitoring programs differ. This guide is for the case where there is nothing to disagree about, because there are no readings at all.

Write the shape down before doing anything else. "All sensor fields blank or zero in GPU-Z and HWiNFO; card name reads correctly; games run normally" is a diagnosis in progress. "The sensors are wrong" is not specific enough to act on, and it will be forgotten by the time Step 2 is reached. The shape is the first piece of evidence, and every later step refers back to it.

"Cannot Read" Is Not the Same as "Reads Zero"

The second distinction to hold onto through the whole guide is between two things that look identical on screen: a sensor that cannot be read, and a sensor that reads zero. They look the same — a 0 in a field — and they mean opposite things.

A reading of zero can be a real measurement. A fan at 0 RPM on a card with a fan-stop feature is a real measurement: the fan is genuinely stopped, by design, because the card is cool enough at idle not to need it. A GPU load of 0% on the desktop is a real measurement: the card genuinely has no work to do. In those cases the instrument is working and the value is zero because the thing being measured is zero.

A reading of zero can also be the way a missing reading is displayed. When the software cannot reach a sensor at all, it has nothing to show, and what it shows instead is often a zero, a blank, or an N/A. A temperature of 0°C is not a measurement of a card at freezing point. It is the display of "no data". A core clock of 0 MHz is not a measurement of a stopped chip. It is the display of "no data". The number is a placeholder for an absence, and reading it as a measurement is the single most common mistake in sensor diagnosis.

The same logic applies to the most famous impossible number in this field: -40°C. A temperature sensor reading -40°C almost always means the sensor cannot be read, not that the card is at minus forty degrees. The value is what the conversion produces when there is no signal to convert. It is a symptom of an unreadable sensor, and it belongs to the temperature-sensor guide, where a single channel is wrong. The same applies to an impossible high reading such as 511°C: like -40°C, it is a value the conversion produces when there is no real signal to convert, and it points at a single sensor layer problem, not at a card running at five hundred degrees. In this guide, the equivalent sign is simpler: every field at once, blank or zero or N/A, while the card demonstrably works. The card is not at zero degrees, zero megahertz, and zero watts simultaneously while running a game. The readings are absent. The question is why the path that carries them has gone dark.

There is a useful rule for telling the two apart. A real zero changes when the thing it measures changes: put the card under load and a real GPU load reading rises; heat the card and a real temperature reading rises; stop the fan-stop feature and a real fan reading rises. A placeholder zero does not change, because there is nothing behind it to change. If the fields stay blank or zero while a game is running and the card is warm and busy, the readings are not measurements. They are absences. That test — load the card and watch whether anything moves — is worth remembering, because it is the fastest way to confirm that the problem is a reporting problem and not a measurement problem.

How a Sensor Reading Reaches the Screen: The Reporting Chain

To find where the readings break, it helps to understand the path they travel. A sensor reading is not measured by the monitoring program. The program does not touch the card. The reading travels through a chain, and if any link in that chain breaks, the reading never arrives. The chain has four levels, and each one is a place the diagnosis can test.

The first level is the sensing level: the physical sensor on the card. Temperature is measured by a diode built into the chip. Fan speed is measured by a tachometer signal from the fan. Power is measured by circuitry on the board. These are physical measurements, made by hardware, and they exist whether or not any software is running.

The second level is the conversion level: the firmware and the analogue-to-digital conversion that turn the physical signal into a number. The diode's electrical signal becomes a temperature in degrees. The fan's pulses become a speed in revolutions per minute. This level lives in the card's firmware, and it is where a raw signal becomes a value that software can use.

The third level is the delivery level: the driver interface that hands the numbers to Windows. For NVIDIA cards this is NVAPI; for AMD cards it is ADL. The monitoring program does not read the sensor. It asks the driver, through one of these interfaces, and the driver answers with the numbers the firmware produced. If the driver is not loaded, or is damaged, or is the generic driver that does not report full telemetry, the numbers stop here — even though the sensors themselves are fine and the card is working.

The fourth level is the display level: the monitoring program itself — GPU-Z, HWiNFO, the manufacturer's own software — which asks the driver for the numbers, receives them, and shows them on screen. The program can only show what the driver gives it. If the program's own core driver fails to load, or the program lacks the permission to reach the driver, or another program is holding the interface it needs, the panel stays blank even though every level below it is working.

There is a parallel chain worth knowing about, because it explains a closely related symptom. The identity of the card — its name, its model, its BIOS version — travels through a different chain from the sensor readings. That chain has four links of its own: enumeration, the driver, the version, and the VBIOS. The two chains are separate, and they can fail separately. That is why the card's name can read correctly while every sensor is blank: the identity chain is intact, and the reporting chain is broken. The identity guide on this site covers the case where the name itself cannot be read. This guide covers the case where the name reads fine and the sensors do not. The two symptoms point in different directions, and the fork between them — identity readable, sensors unreadable — is one of the most useful observations in the whole field.

One more fact about the chain, because it explains a symptom readers find confusing: readings come from the driver and the sensors, not from a direct measurement of the circuit. The monitoring program is reporting what the driver tells it, and the driver is reporting what the firmware tells it. A reading is a report, not a direct observation, and a missing reading is a broken report, not necessarily a broken card. That is why the steps below test the reporting path first, and why the card itself is the last thing to suspect.

The Five Suspicion Layers

A whole panel of missing readings can come from five layers, and they must be kept separate. Mixing them together is how a permissions problem gets treated with a driver reinstall, or a second program holding the interface gets treated as a sensor failure, and how the real cause survives because the wrong layer was treated. The five layers below follow the reporting chain from the monitoring program down to the card, and each one has its own deciding test.

Layer One: The Monitoring Tool's Own Core Driver Layer

The first layer is the monitoring tool itself — specifically, its own core driver. HWiNFO, for example, installs a kernel driver, hwinfo64a.sys, which does the low-level work of reaching the sensors. If that driver fails to install or fails to load, the whole sensor panel shows N/A: temperature, voltage, and fan speed never update, all at once. The cause is not the graphics card. The cause is the tool's own driver, and the failure is total because the tool cannot reach anything without it.

The common triggers are mundane: the tool was not run with administrator permission, the driver file was blocked by Windows, or a Windows update left an old version of the tool unable to load its driver on the new Windows build. The HWiNFO maker's own troubleshooting guide lists "All sensors show N/A" as its first entry, and names exactly these causes: the core driver failing to install or load, no administrator permission, a blocked driver file, another monitoring tool holding the SMBus or embedded-controller interface, and an out-of-date version after a Windows update. The recommended repair follows the same order: close other hardware monitoring programs, run HWiNFO as administrator, restart, and update to the latest version.

The same principle applies to GPU-Z. If GPU-Z recognises the card but the sensors are missing, the problem is usually a driver-layer permission issue or an uncommon sensor exposure path — not a fault in the card. The tool can read the identity and still fail to read the sensors, because identity and telemetry travel on different paths and need different permissions. Running the tool as an ordinary user first, and then as administrator if the sensors fail, is the comparison that tests this layer. If the readings appear under administrator permission, the layer is confirmed, and the answer is about permissions, not about the card.

Layer Two: The Driver Reporting Layer — the Card's Driver Does Not Report Telemetry

The second layer is the graphics driver itself — specifically, the part of it that reports telemetry. The card's name can be read while its telemetry is not reported, because identity and telemetry are two separate paths. The driver can be loaded, the desktop can display, games can run, and still the telemetry path can be broken.

The clearest form of this layer is the generic driver. If Device Manager shows the card as "Microsoft Basic Display Adapter", Windows is running on its generic driver, which provides basic display output and does not report full telemetry. A whole panel of missing readings is the expected result, not a surprise. The card is not broken. The driver that is loaded simply does not do telemetry. The fix is the manufacturer's driver, and Step 5 tests exactly this.

The second form is a damaged driver. The driver is the manufacturer's driver, it loads, the card works — but the telemetry reporting part of it is damaged, and the readings stop. The guide about identity problems notes this shape in passing: if the manufacturer's driver is installed and GPU-Z reads the name but reads no sensors, the fault points at the driver or the sensor layer, and the direction is different from an identity failure. The third form is an old driver: an outdated driver may not recognise a newer card fully, and its telemetry coverage may be incomplete. The manufacturer's own guidance is that the latest driver provides the most complete sensor coverage, and older drivers may report incomplete telemetry. In all three forms, the deciding evidence is the same: the readings are missing everywhere, in every tool, and the driver state — generic, damaged, or outdated — is the common factor.

Layer Three: The Exclusive-Access Layer — Another Program Is Holding the Interface

The third layer is exclusive access: another monitoring program is holding the SMBus or embedded-controller interface and will not let go. Two tools polling the same bus at the same time cannot both read it. The second one gets nothing.

The typical trigger is running several monitoring programs at once: MSI Afterburner, HWiNFO, and the manufacturer's own software, all open together, all polling the same sensors. One of them wins the interface. The others show blank fields, or zeros, or N/A — not because their sensors are broken, but because they cannot get to the bus. The HWiNFO guide names this directly: close other hardware monitoring software first, then test again. It is the cheapest test in the whole guide, it costs nothing, and it should be done before any reinstall, because a reinstall cannot fix a problem that is caused by two programs arguing over one interface.

The deciding evidence for this layer is simple: close every other monitoring program, run one tool alone, and see whether the readings return. If they do, the layer is confirmed, and the answer is to run one monitoring program at a time. If they do not, the layer is ruled out, and the search moves on. Note that this layer and the permissions layer interact: a program run without administrator permission may lose the contest for the interface to a program that has it. That is another reason Step 3 tests permissions and exclusive access together, in one controlled comparison, changing one thing at a time.

Layer Four: The System Security Interception Layer — Memory Integrity and Low-Level Access

The fourth layer is system security: Memory Integrity, or HVCI, restricting low-level access to parts of the embedded controller's registers. This layer can produce a whole panel of N/A readings, and it can also produce a narrower symptom — temperature and power normal, but the fan-speed channel stuck at 0 RPM or N/A. The HWiNFO maker's guide names this as its second entry: "Memory Integrity blocks fan readings", with the cause given as HVCI restricting low-level access to embedded-controller registers, and the repair given as updating HWiNFO to the latest version, checking the Memory Integrity setting, and temporarily switching it off to confirm the cause before switching it back on.

This step is a read-only check. Open Windows Security, go to Device security, open Core isolation details, and look. Do not switch Memory Integrity off as a fix. It is never a fix. At most, switching it off once, confirming whether the readings return, and switching it back on immediately is a diagnostic comparison — and even that comparison comes only after updating the monitoring tool to its latest version, because a current tool version is the supported answer to this layer, and a disabled security feature is not. The same rule that applies to the Code 48 guide applies here: a reading that only appears with a security feature disabled is a reading whose tool or driver should be updated, not a machine that should be left less secure.

A subtlety worth noting: this layer and the tool-version layer overlap. An old version of a monitoring tool may not know how to work with the current Memory Integrity requirements, while a current version does. Updating the tool first, before touching any security setting, is the correct order — and it is the order Step 4 and Step 6 follow.

Layer Five: The Sensor and Firmware Layer — the Card's Own Sensing Hub

The fifth layer is the card itself: the sensor hub on the card, or the sensor table in the VBIOS, not responding. This layer is the last to consider, and it must be earned, not assumed. A single faulty sensor cannot explain a whole panel going dark. Temperature, clocks, power, fan speed, and load are separate sensing points; for all of them to fail at once, the failure has to be in something they share — the sensing hub, or the firmware table that describes them — not in any one sensor. That is the logic that makes this layer the last resort: it is the only layer in which the card itself is the cause, and it can only be claimed after the four layers above it have been tested and cleared.

The deciding evidence for this layer is the cross-test in Step 7. If the card shows the same whole-panel absence on a different machine, with a different driver and different tools, the fault follows the card. If a different, known-good card shows full readings on this machine, the fault does not follow the machine. Only when both halves of that comparison point the same way can the sensor layer be named. Until then, it is a possibility, not a conclusion — and it is never the first possibility.

LayerWhat is wrongDeciding evidenceWhat resolves it
1. Tool's own core driverMonitoring tool's kernel driver fails to install or loadAll sensors N/A in one tool; readings return when run as administratorRun as administrator; update the tool; unblock its driver file
2. Driver reportingGraphics driver does not report telemetry — generic, damaged, or outdatedGeneric driver in Device Manager; readings missing in every toolManufacturer's driver, cleanly installed
3. Exclusive accessAnother monitoring program holds the SMBus or EC interfaceReadings return when only one tool is runningRun one monitoring program at a time
4. System security interceptionMemory Integrity restricts low-level EC register accessReadings missing or fan channel only affected; current tool version helpsUpdate the tool; read-only check only — never leave security off
5. Sensor / firmwareCard's sensor hub or VBIOS sensor table not respondingSame whole-panel absence on a second machineHardware assessment, only after all other layers are cleared

The Seven Diagnostic Steps

The seven steps below are ordered so that the cheapest, least destructive tests come first, and so that no step destroys the evidence a later step needs. Work through them in order, one variable at a time, and write down the result of each step before moving on. Changing two things at once and seeing the readings return leaves the reader not knowing which change mattered — and not knowing what to do when the readings disappear again.

Step 1: Confirm the Shape of the Symptom

Before touching anything, write down the shape of the symptom. The card's name reads correctly in GPU-Z — write the name down. The display works, games run — note that too. Then list every sensor field, and what it shows: temperature blank, core clock 0 MHz, memory clock 0 MHz, power draw 0 W, fan speed 0 RPM or N/A, GPU load 0%. Be specific. "All sensors blank or zero" is a claim. A list of fields is evidence.

This step proves the symptom has been narrowed to the reporting path as a whole: identity reads, telemetry does not, and the card demonstrably works. It proves nothing about which layer broke — that is what the steps below are for. But the shape is the fork. If the list shows that only one field is wrong — only the temperature, only the fan speed, only the load — then this guide does not apply, and the reader should follow the guide for that single channel: the temperature-sensor guide, the fan-speed guide, or the utilisation guide. If the list shows every field missing together, this guide applies, and the search is for a shared failure in the reporting path.

Note also whether the fields are blank, zero, or N/A, and whether they change under load. Open a game, let the card warm up and work, and watch the panel. If every field stays blank or zero while the card is busy, the readings are absences, not measurements — a real zero would move when the thing it measures moves. Write that observation down too. It is the fastest confirmation that the problem is a reporting problem, and it costs nothing.

Step 2: Cross-Check with a Second Tool

Open a second monitoring tool and compare. GPU-Z against HWiNFO, and against the manufacturer's own software — NVIDIA App or AMD Software — gives three sides to compare. Do not install a new tool for this step if two are already available; use what is already on the machine. The comparison is the test, and it costs nothing and requires no reinstall.

If all three tools agree — every one of them shows the same blank or zero panel — the fault is below the tools, in a layer they share: the driver reporting layer, or the sensor layer. The tools disagree with nothing, because there is nothing to disagree about, and the shared absence points downward. If the tools disagree — one shows readings and another does not — the fault is in the software layer: the tool that shows nothing has a problem of its own, in its core driver, its permissions, or its access to the interface. The disagreement is itself the diagnosis, and it points at the tool, not at the card.

This step belongs before any reinstall, because it is free and it separates a tool problem from a driver problem without changing anything. A driver reinstall cannot fix a tool that has lost a contest for the interface, and a tool reinstall cannot fix a driver that reports no telemetry. The cross-check tells the reader which one to reinstall — if a reinstall is needed at all — and it should never be skipped.

Step 3: Permissions and Exclusive Access — One Tool, as Administrator

Close every monitoring program. All of them: Afterburner, HWiNFO, GPU-Z, the manufacturer's software, any hardware monitor running in the background or in the system tray. Then open one tool only — one — and run it as administrator: right-click the program and choose "Run as administrator". Look at the sensor panel again.

If the readings return, the first or third layer is confirmed: the problem was permissions, or exclusive access, or both. The tool needed administrator permission to reach its core driver, or it needed the interface to itself without another program holding it. The answer is to run one monitoring program at a time, with the permission it needs. Write down which change mattered — the permission, or the closing of the other programs — by testing them separately if necessary: first one tool alone as an ordinary user, then the same tool alone as administrator. One variable at a time. That is how the reader learns which one it was.

If the readings stay missing, the first and third layers are ruled out, and the search moves on. The tool has the permission it needs and the interface to itself, and still there is nothing to show. That is useful information: it means the fault is below the tool, in the driver or the sensor layer, and the next steps test those.

A note on what this step is not: it is not a reason to run everything as administrator permanently, and it is not a reason to disable security features. Administrator permission for a monitoring tool is a normal requirement for a program that reads low-level hardware interfaces. It is not a workaround, and it is not a security downgrade. The distinction matters, and it is worth keeping straight.

Step 4: Update the Monitoring Tool to the Latest Version

Update GPU-Z and HWiNFO to their latest versions. An old version may not recognise a newer card, or a newer Windows build, and its sensor coverage may be incomplete. The tool makers' own guidance is consistent on this point: the latest driver — and the latest tool version — provides the most complete sensor coverage, and older versions may report incomplete telemetry. A Windows update can also leave an old tool version unable to load its core driver on the new build, which produces exactly the whole-panel N/A symptom this guide covers.

This step belongs before any driver reinstall, because it is cheaper and because it tests a different layer. Updating the tool tests the tool layer. Reinstalling the graphics driver tests the driver layer. Doing the driver first and the tool second wastes the evidence of which one mattered. Update the tool, restart it, and look at the panel. If the readings return, the version was the problem, and the diagnosis is done. If they do not, the tool layer is cleared, and the search moves to the driver.

Note the version numbers before and after, next to the Step 1 note. "HWiNFO 8.52, readings returned" is a diagnosis. "Updated the tool" is not specific enough to be one.

Step 5: Check the Driver State — Generic, Manufacturer's, or Damaged

Open Device Manager, expand Display adapters, and look at the driver provider. Is it the manufacturer — NVIDIA, AMD, or Intel — or is it Microsoft? If the card shows as "Microsoft Basic Display Adapter", Windows is running on its generic driver, which provides basic display output and does not report full telemetry. A whole panel of missing readings is the expected result. The card is not broken. The driver that is loaded simply does not do telemetry. Install the manufacturer's driver, and check the panel again.

If the manufacturer's driver is already installed and the whole panel is still blank or zero, the driver reporting layer is the suspect — and the direction is a clean reinstall of the driver, not a repair of the card. A driver can load, display the desktop, and run games, while the telemetry reporting part of it is damaged. The guide about identity problems notes this shape: if the manufacturer's driver is installed and GPU-Z reads the name but reads no sensors, the fault points at the sensor layer or a damaged driver, and the direction is different from an identity failure. The clean-install procedure — removing the old driver fully in Safe Mode with Display Driver Uninstaller, then installing the manufacturer's driver — belongs to its own guide; this guide says only that a clean removal followed by a clean installation is the correct form of "reinstall the driver", and that installing a new driver over the top of an old one is not the same thing.

If the readings return after the clean install, the driver layer is confirmed. If they do not, the driver layer is cleared — the driver is current, cleanly installed, and the manufacturer's own — and the remaining suspects are the security layer and the sensor layer, which the next two steps test.

Step 6: Read-Only Check of Memory Integrity

Open Windows Security, go to Device security, and open Core isolation details. Look at the Memory Integrity setting and at whether anything is listed there as incompatible. This step is a read-only check. It teaches the reader how to look, not how to switch anything off.

Memory Integrity can restrict low-level access to embedded-controller registers, and that restriction can produce a whole panel of N/A readings — or a narrower symptom, with temperature and power normal but the fan-speed channel stuck at 0 RPM or N/A. The tool maker's guidance for that narrower symptom is to update the monitoring tool to the latest version first, check the Memory Integrity setting, and only then consider temporarily switching it off to confirm the cause before switching it back on. That order matters: the tool update comes first, because a current tool version is the supported answer, and a disabled security feature is not.

If the readings are missing and Memory Integrity is on, note it, and note whether updating the tool in Step 4 already changed anything. Do not switch Memory Integrity off as a fix. Do not leave it off. A reading that only appears with a security feature disabled is a reading whose tool or driver should be updated, not a machine that should be left less secure. And never disable driver signature enforcement for a sensor problem: it has nothing to do with sensor reporting, and it removes a protection against malicious drivers. If a source offers either of those as a fix for blank sensors, close it.

Step 7: Cross-Test, and Know the Stop Line

The last step is a cross-test, and it is the step that separates a machine problem from a card problem. Either test the card in a different Windows computer, with the manufacturer's driver and a current monitoring tool, or test a different, known-good card in this computer. One variable changes. Everything else stays the same.

If the card shows the same whole-panel absence on the second machine, the fault follows the card: the sensor hub or the VBIOS sensor table is not responding, and the sensor layer is the answer. If the card shows full readings on the second machine, the fault follows this machine: the driver, the tool, or the security layer on this machine is the problem, and the search returns to Steps 3 to 6 with the knowledge that the card itself can report. If a different, known-good card shows full readings on this machine, the machine's reporting path is healthy and the fault follows the original card. Each outcome points somewhere different, and each one is worth writing down, because it is the answer the next person to look at the machine will need.

The stop line is this: if the whole panel is missing and, in addition, Device Manager cannot recognise the card, or the machine shows no picture at boot, or there is a smell of burning, stop working in the software layers and bring the machine in. Those are not reporting problems. They are hardware problems, and no further software comparison will help. Bring the machine in with the notes from Steps 1 to 7: the list of blank fields, the cross-tool comparison, the permissions and exclusive-access result, the tool versions before and after, the driver provider and version, the Core isolation details result, and the cross-test outcome. Inspection takes place at Room 1209, Dali Plaza, Shek Mun. You can call 2186 8010, or reach the shop on WhatsApp at 6558 6806. Inspection is charged at the site's published three-tier fees of HK$400–2,800, quoted after inspection. There is no home-visit service and no Mong Kok shop; inspection is carried out in the shop.

StepWhat to doWhat it provesWhat it does not prove
1. Confirm the shapeList every sensor field; note card name reads correctly; load the card and watchThe symptom is the whole panel at once, not one channelWhich layer broke — that comes later
2. Cross-check with a second toolGPU-Z vs HWiNFO vs manufacturer's softwareWhether the fault is in one tool or in a shared layer belowAnything, if only one tool was tried
3. Permissions and exclusive accessClose all monitors; run one tool alone, as administratorWhether permissions or exclusive access is the causeAnything about the driver or sensor layers
4. Update the monitoring toolUpdate GPU-Z / HWiNFO to the latest versionWhether the tool version was the problemAnything about the driver
5. Check the driver stateDevice Manager: manufacturer or Microsoft provider; clean reinstall if neededWhether the driver reports telemetry at allAnything about the security or sensor layers
6. Read-only Memory Integrity checkWindows Security, Device security, Core isolation details — look onlyWhether security interception is involvedNothing, if the tool has not been updated first
7. Cross-test and stopSame card in another machine, or a known-good card in this oneWhether the fault follows the card or the machineNothing beyond the test that was actually run

Six Common Misconceptions

Misconception 1: Readings all zero means the graphics card is broken

No. A whole panel of missing readings points at the reporting path as a whole — the tool's core driver, permissions, exclusive access, security interception, or the graphics driver — not at the card itself. The card has a picture, runs games, and reads its name correctly. That is strong evidence against a hardware fault. The card itself is the last layer to consider, and only after the four layers above it have been tested and cleared, and only on the evidence of a cross-test. A blank sensor panel, on its own, is never evidence that the card is faulty.

Misconception 2: A temperature of 0°C means the card is at freezing point

No. A reading of zero is not always a measurement. When the software cannot reach a sensor at all, what it shows is often a zero, a blank, or an N/A — and that zero is the display of "no data", not a measurement of zero degrees. A card cannot be at 0°C, 0 MHz, and 0 W simultaneously while running a game. The value is a placeholder for an absence. The same applies to -40°C, which usually means the sensor cannot be read, not that the card is at minus forty degrees. The test is whether the value moves when the thing it measures changes: a real zero moves under load, and a placeholder zero does not.

Misconception 3: A fan speed of 0 RPM means the fan is broken

No. On a card with a fan-stop feature, 0 RPM at idle is normal design: the fan is genuinely stopped because the card is cool enough not to need it. The question is not whether the fan turns at idle. The question is whether it turns when the card gets hot. A fan that stays still under load is a different symptom from a fan that is stopped at idle, and the two should never be confused. A single fan-speed channel reading 0 RPM while the fan is visibly turning belongs to the fan-speed guide, where the reporting wire is the suspect — not to this guide, and not to a conclusion about the fan.

Misconception 4: If two programs show different numbers, one of them must be broken

No. Sampling points, polling intervals, and definitions differ between programs, so numbers are expected to differ. Board power and chip power are different measurements. A temperature sampled every second and a temperature sampled every half-second will disagree at any given moment. The correct comparison is of trends, not of absolute values: do both programs rise together under load and fall together at idle? If they do, both are working, and the difference is in what each one measures. That is the guide about why monitoring programs differ — and it is about readings that exist, not about readings that are missing.

Misconception 5: A missing sensor field means that sensor is broken

No. Not every model has every sensing point. Some models have no hotspot reading — the manufacturer hides it in the official interface, and that is normal, not a fault. Some cards have no memory-junction temperature sensor, and the software hides the field, leaves it blank, or substitutes a reading from another location. A missing field, on its own, is not a fault. A whole panel missing at once is a different shape, and it points at the reporting path, not at any single sensor. The distinction between one absent field and every field absent is the distinction this guide is built on.

Misconception 6: Normal sensor readings mean the card is healthy

No. Readings only reflect the links the sensors can see. An intermittent fault, a problem in an individual memory chip, or a driver problem may not show up in temperature or power draw at all. A card can read perfectly normally and still be faulty, and a card can show abnormal readings and still be healthy — because a reading is evidence, not a verdict. The reverse is equally true: abnormal readings do not, on their own, prove the card is faulty. Readings are one piece of evidence among several, and they must be read together with the card's behaviour — whether games run, whether the display is stable, whether the card is recognised — before any conclusion is drawn.

Boundaries: Problems That Look Similar but Are Not This One

Not every sensor problem is a whole-panel problem, and the single-channel problems each belong to their own guide. If the symptom is one of the following, follow that guide, not this one.

A single temperature channel reading wrong — -40°C, a value stuck at one number, a value jumping wildly — belongs to the temperature-sensor guide, which explains the four-level reading chain from the on-die diode through firmware conversion and the driver interface to the display software. A single fan-speed channel reading 0 RPM while the fan is visibly turning belongs to the fan-speed guide: the drive path and the reporting path are two separate paths, and the speed-reporting wire can fail while the fan keeps turning. A single utilisation reading of 0% while a game runs belongs to the utilisation guide: the monitor may be showing the busiest engine rather than the average, or the wrong card may have been selected. An identity that cannot be read at all — the card's name showing as Unknown, or the model field blank — belongs to the identity guide, which explains the four-link reading chain of enumeration, driver, version, and VBIOS. Readings that exist but disagree between two programs belong to the guide about why monitoring programs differ: sampling points and definitions, not missing data.

Three more boundaries deserve a sentence each. The first is the temperature reading that disappears after waking from sleep: if only the temperature is missing, and only after waking, that belongs to the temperature-sensor guide, where a driver bug is the suspect. If the whole panel disappears after waking, that belongs to this guide, in the driver reporting layer. The second is the missing hotspot reading on newer cards: the manufacturer hides it in the official interface, and that is normal, not a fault — do not go looking for why it is gone. The third is the idle fan at 0 RPM: fan-stop is a normal feature, not a fault, and it is covered by its own guide.

Finally, a reading problem accompanied by crashing, black screens, or other hardware symptoms belongs to a different direction entirely. This guide is for a reporting problem on a machine that otherwise works. When the machine does not work, the reporting problem is not the problem to solve first.

When to Stop and Bring the Machine In

Stop when all seven steps have been worked through, one variable at a time, and the whole panel is still blank or zero. Stop sooner if the card disappears from Device Manager, if the machine shows no picture at boot, if there is a smell of burning, or if any step involves deleting system files or registry entries. Stop, above all, when the cross-test points at the card itself — the same whole-panel absence on a second machine — because that is the point where home diagnosis has done its job and the fault is in the card's own sensing or firmware.

Bring the machine in with the notes from every step: the list of blank fields, the cross-tool comparison, the permissions and exclusive-access result, the tool versions before and after, the driver provider and version, the Core isolation details result, and the cross-test outcome. In-store inspection takes place at Room 1209, Dali Plaza, Shek Mun. You can call 2186 8010, or reach the shop on WhatsApp at 6558 6806. Inspection is charged at the site's published three-tier fees of HK$400–2,800, quoted after inspection. There is no home-visit service and no Mong Kok shop; inspection is carried out in the shop.

A diagnosis that ends at the stop line is not a failed diagnosis. A note that says "Card name reads correctly; all sensor fields blank or zero in GPU-Z and HWiNFO; readings return in no tool; one tool alone as administrator, still blank; tools updated, still blank; manufacturer's driver cleanly installed, still blank; same card in a second machine, whole panel blank" is a precise description of where the fault lives, and it is exactly what an in-store inspection needs to start from the right place instead of from the beginning.

FAQ

Q1: All my GPU-Z sensor readings are blank or zero. Does that mean my graphics card is broken?

No. A whole panel of missing readings points at the reporting path as a whole — the tool's core driver, permissions, exclusive access, security interception, or the graphics driver — not at the card itself. The card has a picture, runs games, and reads its name correctly, which is strong evidence against a hardware fault. The card itself is the last layer to consider, and only after the four layers above it have been tested and cleared, and only on the evidence of a cross-test.

Q2: What is the difference between the whole panel going blank and one sensor reading being wrong?

The whole panel at once means a shared reporting path has failed: temperature, clocks, power, fan speed, and load all travel the same path, and when it breaks, everything on it goes dark together. One channel alone means that channel is at fault: a single temperature channel reading -40°C belongs to the temperature-sensor guide, a single fan-speed channel reading 0 RPM while the fan turns belongs to the fan-speed guide, and a single load reading of 0% belongs to the utilisation guide. The shape of the failure decides which guide applies.

Q3: How do I tell "cannot read" apart from "reads zero"?

"Cannot read" means the tool cannot get the data: the field shows 0, N/A, or blank, and that is the display of "no data", not a measurement. "Reads zero" means the sensor genuinely measured zero — for example, a fan genuinely stopped at idle really is 0 RPM. The test is the shape: every field at 0 or N/A together points at "cannot read"; a single field at 0 while the other fields are normal, and the real behaviour matches — the fan really is stopped at idle — points at a real zero.

Q4: My temperature reads 0°C. Is my card really at freezing point?

No. A reading of zero is not always a measurement. When the software cannot reach a sensor at all, what it shows is often a zero, a blank, or an N/A — and that zero is the display of "no data", not a measurement of zero degrees. A card cannot be at 0°C, 0 MHz, and 0 W simultaneously while running a game. The test is whether the value moves when the thing it measures changes: a real zero moves under load, and a placeholder zero does not. The same applies to -40°C, which usually means the sensor cannot be read.

Q5: My fan speed reads 0 RPM. Is the fan broken?

Not necessarily. On a card with a fan-stop feature, 0 RPM at idle is normal design: the fan is genuinely stopped because the card is cool enough not to need it. The question is whether the fan turns when the card gets hot, not whether it turns at idle. If the whole panel is blank as well, the 0 RPM is most likely a missing reading, not a measurement of a stopped fan. A single fan-speed channel reading 0 RPM while the fan is visibly turning belongs to the fan-speed guide.

Q6: GPU-Z shows nothing but HWiNFO shows readings. Which one is right?

The disagreement itself is the diagnosis: the fault is in the software layer, in the tool that shows nothing. Its core driver, its permissions, or its access to the interface is the problem — not the card. Close every other monitoring program, run the failing tool alone as administrator, and update it to the latest version. If the readings return, the tool layer was the cause. If all tools agree that the readings are missing, the fault is below the tools, in a layer they share.

Q7: GPU-Z and HWiNFO show different numbers. Is one of them broken?

No. Sampling points, polling intervals, and the board's definition of chip power differ, so the numbers are supposed to differ. Two programs both having numbers that differ is a sampling and definition difference, not one of them being broken. Compare the trend first, then the value: if the temperature in both programs rises and falls together, that is normal. This is different from the whole panel having no readings at all: numbers that differ belong to the guide about why monitoring programs differ; no readings at all belongs to this guide.

Q8: Will turning off Memory Integrity bring my sensor readings back?

It might, but it must never be left off as a fix. Memory Integrity can restrict low-level access to embedded-controller registers, which can block sensor readings. The supported answer is to update the monitoring tool to the latest version first, because a current tool version is designed to work with the current security requirements. Switching Memory Integrity off is, at most, a diagnostic comparison done once and switched back on immediately. Never disable driver signature enforcement for a sensor problem: it has nothing to do with sensor reporting.

Q9: Device Manager shows "Microsoft Basic Display Adapter". Is that why my sensors are blank — and what if only one sensor field is missing?

Yes, most likely. The generic driver provides basic display output and does not report full telemetry, so a whole panel of missing readings is the expected result. The card is not broken. The driver that is loaded simply does not do telemetry. Install the manufacturer's driver for the exact card model and Windows version, cleanly, and check the sensor panel again. If the readings return, the driver layer was the cause.

Not necessarily. Not every model has every sensing point. Some models have no hotspot reading — the manufacturer hides it in the official interface, and that is normal. Some cards have no memory-junction temperature sensor, and the software hides the field or leaves it blank. A missing field, on its own, is not a fault. What matters is the shape: one field missing is usually normal or a single-channel issue, while every field missing together points at the reporting path, and that is what this guide covers.

Q10: When should I stop trying things myself and bring the machine in for inspection?

Stop when all seven steps have been worked through, one variable at a time, and the whole panel is still blank or zero; or when the cross-test shows the same whole-panel absence on a second machine; or sooner if Device Manager cannot recognise the card, the machine shows no picture at boot, or there is a smell of burning. Bring your notes: the list of blank fields, the cross-tool comparison, the permissions and exclusive-access result, the tool versions before and after, the driver provider and version, the Core isolation details result, and the cross-test outcome. In-store inspection is at Room 1209, Dali Plaza, Shek Mun, charged at the site's published three-tier fees of HK$400–2,800, quoted after inspection. You can call 2186 8010 or WhatsApp 6558 6806.

Related articles

Related services

If you need Chip-level GPU repair, see these service pages: