How to Read GPU-Z / HWiNFO Sensor Data to Diagnose GPU Faults
Key Takeaways
Monitoring data is a first step in diagnosis, not a verdict. The useful habit is to compare idle and loaded readings, look for several readings pointing the same way, and capture the numbers at the moment of failure. This guide explains each GPU-Z and HWiNFO reading, what abnormal patterns mean, and the misreadings that lead to wrong conclusions.
Introduction: A Wall of Numbers, and No Idea Which One Matters
A graphics card seems unwell. The advice everywhere is the same: "open GPU-Z and have a look", or "check HWiNFO". So the tools are opened — and there they are: temperatures, clocks, loads, power figures, fan speeds, and a column of cryptic labels. Which of these numbers matters? Which one is abnormal? What does it mean when one is high and another is low?
This guide answers those questions. It covers what each key reading means, how to read the readings together rather than one at a time, and the common misreadings that send people chasing the wrong fault. No disassembly is required, and no money needs to be spent: monitoring is the cheapest first step in graphics card diagnosis, and the records it produces are genuinely useful to a repair shop later.
What GPU-Z and HWiNFO Each Do
| Tool | Scope | Best for |
|---|---|---|
| GPU-Z | Graphics-card-specific information and monitoring: model, BIOS version, memory capacity, live sensor readings | Card-focused checks: temperatures, clocks, power, fan, PerfCap |
| HWiNFO | Whole-system monitoring: graphics card plus CPU, motherboard, and other sensors | Cross-checking the card against the rest of the system, and logging over time |
One point about the nature of these readings: GPU-Z and HWiNFO report what the drivers and on-board sensors report. They do not measure the board's circuitry directly. A sensor can itself be faulty, a reading can lag behind reality, and some faults — an intermittent memory chip fault, for example — may not show up in any sensor reading at all. Sensor data is evidence, not a verdict.
Each Key Reading, Explained
| Reading | What it is | What an abnormal pattern tends to suggest |
|---|---|---|
| GPU Temperature | The temperature reported for the GPU die, typically at the edge or an averaged point | High under load with falling clocks tends to suggest a cooling or thermal-limit problem |
| Hotspot temperature | The hottest point measured on the die, reported by a separate sensor | A large gap between hotspot and GPU temperature tends to suggest uneven contact — paste, pads, or mounting pressure; both high together tends to suggest overall cooling insufficiency |
| Memory temperature | Temperature of the GDDR memory chips | High memory temperature with a normal core temperature tends to suggest a memory-cooling problem (thermal pads, memory heatsinking) |
| GPU Clock / Memory Clock | The current operating frequencies of core and memory | Clocks far below their usual loaded values, or swinging wildly under steady load, tend to suggest throttling from temperature, power, or supply limits |
| GPU Load | How busy the GPU is | Load that cannot rise, or load high while clocks stay low, is a clue that something is limiting the card |
| Power Draw / Board Power | Actual power being drawn | Compare against the power limit: drawing at the limit with low clocks points at a limit or supply issue; drawing far below it during a heavy load points elsewhere |
| PerfCap Reason (NVIDIA) | The reason performance is currently limited: Pwr (power limit), Thrm (temperature limit), VRel (voltage reliability limit), VOp (voltage ceiling), Util (utilisation), Idle (standby) | A limit flag at full load can be normal behaviour, not a fault — see the misreadings section |
| Fan Speed (RPM and percentage) | How fast the fans are spinning | High temperature with low or zero fan speed points toward a fan-control or fan fault; note fan-stop behaviour below |
A few of these deserve a longer explanation.
Hotspot versus GPU temperature. These are two different sensors. The GPU temperature is the headline figure; the hotspot is the hottest point on the die. Some gap between them is normal, because heat is never perfectly uniform. What matters is the pattern: a wide gap with a cool headline temperature tends to point at uneven contact between the cooler and the die — dried paste, a poorly seated cooler, or uneven mounting pressure. Both readings high together tends to point at the cooling system as a whole — airflow, a clogged heatsink, or a cooler that cannot shed the heat. This guide deliberately gives no universal "normal gap" figure, because cooler designs differ between cards; the right comparison is the same card's behaviour now versus its behaviour when it was healthy.
PerfCap Reason. On NVIDIA cards, GPU-Z shows a PerfCap Reason field that names the current performance limiter. Pwr means the card is at its power limit. Thrm means it is at its temperature limit. VRel and VOp are voltage-related limits. Util means the card is simply not being fully used, and Idle means it is at rest. The critical point: a limit flag at full load is often normal operation. A card running a heavy workload is expected to sit at one of its limits — that is what the limits are for. PerfCap becomes diagnostic only in combination: high temperature plus falling clocks plus Thrm points toward overheating; a power limit flag with low clocks and normal temperatures points toward a power or supply question.
Fan speed and fan-stop. Many modern cards have a fan-stop feature: the fans stay off at low temperatures and start only when needed. Zero fan speed at idle is normal behaviour on such cards, not a fault. The fault pattern is the opposite: high temperature with fans still slow or stopped, or a fan that never spins even under heavy load.
How to Use the Readings: Four Working Methods
Compare trends, not single points. A single reading means little. Record two sets: one at idle (desktop, no workload), and one under a sustained load. The change between them is where the information lives. A card that idles normally and collapses under load tells a different story from a card that is already hot at idle.
Cross-confirm with multiple readings. A diagnosis needs at least two or three readings pointing the same way:
- High temperature plus falling clocks plus a Thrm PerfCap flag: overheating and throttling, pointing at cooling.
- Normal temperature plus low clocks plus a Pwr flag at full load: the power limit is the active constraint, which may be normal — or a supply question if performance has changed.
- High memory temperature plus normal core temperature plus artifacts: the memory cooling path deserves attention.
- High temperature plus low or zero fan speed: the fan or its control is suspect.
A single high number, on its own, proves nothing. Temperature readings in particular must be read alongside clocks, load, and fan speed before any conclusion is drawn.
Capture the moment of failure. Readings taken after a crash, when the card has returned to idle, describe a healthy idle card — not the failure. Use the logging feature in GPU-Z or HWiNFO to record readings continuously, so the seconds before and during the failure are preserved. That log is the most valuable piece of evidence a user can hand to a repair shop: it shows what the card was doing when it failed, not what it does when it is fine.
Compare the card with itself. Every card has a different cooler design, so there is no single "normal temperature" that applies to all cards. The meaningful comparison is the same card now versus the same card when it was healthy: temperatures higher than they used to be under the same workload, clocks lower than they used to be, or fan behaviour that has changed. If no personal baseline exists, the change over time — before and after cleaning, for instance — is still informative.
Common Misreadings
- Reading a single number. A temperature figure alone, without clocks, load, and fan speed, cannot diagnose anything.
- Applying a universal temperature threshold. Cooler designs differ widely. A temperature that is ordinary for one card may be high for another. There is no single threshold that separates healthy from faulty across all cards.
- Treating PerfCap flags as faults. Pwr or Thrm at full load is usually the card operating at its designed limits. The flag matters only when combined with changed behaviour — lower performance than the card used to deliver, or instability.
- Mistaking fan-stop for a dead fan. Fans that stay off at idle on a card with a fan-stop feature are behaving as designed. Judge fan behaviour under load, not at the desktop.
- Assuming normal readings mean a healthy card. Sensors cannot see every fault. An intermittent memory fault, a failing solder joint, or a VBIOS problem can all exist alongside perfectly ordinary sensor data. Normal readings narrow the possibilities; they do not clear the card.
What to Bring to a Repair
A short monitoring record makes a repair diagnosis faster and more accurate. Before arranging a repair, capture:
- Idle readings and loaded readings (temperature, hotspot, memory temperature, clocks, load, power, fan speed).
- The log covering the moment of failure, if the failure can be reproduced.
- A plain description of what changed: what the card used to do under the same workload, and what it does now.
- Whether the symptoms track temperature, load, or neither.
That record, plus the elimination tests from the other guides in this series, gives the repair shop a starting point grounded in evidence rather than guesswork.
FAQ
Q: My hotspot reading is much higher than my GPU temperature. Is something wrong?
A: A gap between the two readings is expected, since the hotspot is the hottest point on the die. A wide gap with a cool headline temperature tends to point toward uneven contact — paste, pads, or mounting pressure. Both readings high together tends to point toward overall cooling. The useful comparison is the same card's gap now versus its gap when healthy, not any universal threshold.
Q: PerfCap shows Pwr. Does that mean the PSU is too weak?
A: Not necessarily. Pwr means the card is operating at its power limit, which is normal behaviour under heavy load — the limit is doing its job. It becomes a question about power delivery only when combined with other evidence: performance that has dropped from what the card used to deliver, instability under load, or power readings that do not match the workload. A Pwr flag alone is not a PSU diagnosis.
Q: My fans do not spin at idle. Are they broken?
A: Probably not. Many cards stop their fans at low temperatures by design (fan-stop). Check fan behaviour under load: fans that stay slow or stopped while temperatures climb are the fault pattern. Fans that are off at idle and spin up under load are behaving normally.
Q: Can sensor readings prove a card is healthy?
A: No. Sensors report what they can measure: temperatures, clocks, power, fan speed. Faults that do not change those readings — an intermittent memory chip fault, a failing solder joint, a VBIOS problem — can exist alongside normal sensor data. Normal readings narrow the possibilities; they never clear a card completely.
Q: HWiNFO and GPU-Z show different temperatures. Which one is right?
A: Both tools read the same sensors through the driver, so large differences usually come from reading at different moments or from different sensor selections (one showing GPU temperature, the other showing hotspot). Compare like with like — the same sensor, at the same load, at the same time — before concluding that a tool is wrong.
Q: How do I use the logging feature?
A: Both tools can log readings over time. Start logging while reproducing the problem, so the readings from before and during the failure are captured. Hand the log file to the repair shop together with a description of the problem, so they can see what happened at the moment of failure.