Your GPU hit 105 °C on the hotspot, the screen went black mid-raid, and now you're staring at a case that sounds like a leaf blower. You're not alone. But before you order a new graphics card, know this: the overwhelming majority of GPU overheating problems are cooling problems, not dead-silicon problems. I've fixed this on everything from a dust-choked GTX 1080 to a waterblocked card that had a mounting-pressure issue nobody spotted for weeks, and I'll walk you through exactly how to troubleshoot it.
Here's the thing most guides skip: your GPU doesn't have one temperature. It has several, and the one that actually triggers a shutdown isn't the one your overlay shows by default. Get that wrong and you'll spend hours cleaning a cooler that was never the culprit.
Key Takeaways
- 70 °C under load is normal for most modern GPUs. You're not in trouble until you cross specific higher thresholds.
- Your overlay shows the edge temperature, not the hotspot. The hotspot runs 10-25 °C higher and is the number that matters.
- A hotspot delta above 20-25 °C points to a contact problem (thermal paste, mounting pressure), not airflow.
- Custom fan curves and a mild undervolt are free fixes that solve a shocking number of cases before you touch a screwdriver.
- Dust is the cause people assume is number one. It isn't always. Thermal paste that has pumped out after three years is the stealth killer.
- If temperatures are fine but the card still crashes, you're troubleshooting the wrong thing.
How to troubleshoot GPU overheating without wasting a weekend
The mistake I made early on was treating every overheat as a dust problem. I'd blow out the card, bolt it back in, and feel smug for about forty-five minutes until the shutdowns returned. The card wasn't dusty. The paste had dried into a chalky crust, and no amount of compressed air was going to fix that.
So here's the order that actually works. Follow it top to bottom, and stop the moment you find your culprit.
Step 1: measure before you touch anything
Install a monitoring tool that reads the hotspot sensor. On most cards, GPU-Z and HWiNFO both expose it; HWiNFO labels it "GPU Hotspot Temperature" and you want to run it in Sensors-only mode so it doesn't fight with your other tools. Run a real load for ten to fifteen minutes. A game works. A stress test works better because it's repeatable.
Write down four numbers: idle edge temp, load edge temp, load hotspot temp, and fan RPM at that load. Those four figures tell you almost everything before you open the case.
What do the numbers mean?
- Edge under 75 °C, hotspot under 90 °C, delta under 15 °C — your cooling is fine. Look elsewhere for the crash.
- Edge around 80-85 °C, hotspot climbing past 100 °C, delta 20 °C or more — contact problem. Repaste or recheck the mount.
- Edge and hotspot both maxed, fans screaming at 100% — airflow or ambient temperature issue.
- Fans barely spinning while temps climb — fan curve, fan failure, or a card that thinks it's idle.
Step 2: clean it, then prove the airflow is real
Power down, unplug, and pull the card. Compressed air through the fins, holding the fans still so you don't overspin the bearings. I've seen a single mat of cat hair drop load temps by 8 °C on a friend's build. But don't stop at the card. Look at your case.
Here's the test that settled an argument I'd been having with myself for weeks: run the PC with the side panel off. If temperatures drop more than 5 °C, your case airflow is the problem, full stop. Intake and exhaust need to be balanced. Three intake fans and zero exhaust just circulates hot air in a box.
Step 3: try the free software fixes before you disassemble
MSI Afterburner is the tool I keep coming back to, and its fan curve editor is where most people leave easy wins on the table. The default curve is conservative. It keeps the card quiet at the cost of heat. Drag that curve up so fans ramp harder earlier, and re-test.
Then undervolt. This is the move I wish I'd learned years sooner. Cutting core voltage by 50-100 mV while keeping clocks steady dropped my own card's hotspot by 11 °C with zero performance loss in the games I tested. You also cut power draw, which cuts heat. Two birds, no money spent.
If you're on a laptop, you have one more lever: a cooling pad and, more importantly, elevating the rear of the chassis. Laptop intakes are usually on the bottom. Sitting flat on a bed or a lap starves them.
What are the common causes of a GPU overheating?
Most overheating traces back to a short list, and they behave differently. Knowing which one you have saves you from fixing the wrong thing.
| Cause | Tell-tale sign | Fix | Effort |
|---|---|---|---|
| Dust and debris | High edge and hotspot, both elevated evenly | Compressed air, clean fins | Low |
| Dried or pumped-out paste | Hotspot delta above 20-25 °C | Repaste, remount | Medium |
| Poor case airflow | Temps drop with side panel off | Balance intake/exhaust | Low |
| Aggressive default fan curve | Fans quiet, temps high under load | Custom curve in Afterburner | Low |
| Weak mounting pressure (waterblock, aftermarket cooler) | One corner of the die runs hot | Re-torque screws evenly | Medium |
| High ambient temperature | Everything runs hot, including idle | Improve room cooling | Low |
Notice what's missing from the top of that list: a "dying GPU." Genuine hardware failure on a card that used to run cool and suddenly doesn't is rare. What's common is thermal paste that slowly pumped out over three years, and what's rare is a card that fails without warning. The gradual stuff is what you're hunting.
Is 70 C too hot for a GPU?
No. 70 °C under load is a healthy number for almost any modern graphics card, and depending on the model, it might even be on the cool side. It will not damage anything, and it will not throttle you.
Here's how to read the range. Idle temperatures in a normal room sit somewhere between 30 and 50 °C. Under load, edge temps of 60-80 °C are completely expected, and many cards are designed to sit there happily for their entire service life. The hotspot can run 10-25 °C above the edge without raising an eyebrow. Manufacturers set throttle points well above these numbers precisely so you get performance, not silence, first.
The number that should actually make you sit up is the hotspot crossing roughly 100-105 °C, or an edge temperature in the high 80s and climbing. That's where throttling gets aggressive and shutdowns begin. So if your card reads 70 °C, close the monitoring tool and go play something.
Why the hotspot is the number that matters
A GPU die isn't one uniform slab. The hotspot sensor reports the single hottest point on it, which is usually a specific block of cores working hardest. If your edge reads 72 °C and your hotspot reads 78 °C, you're golden. If your edge reads 72 °C and your hotspot reads 99 °C, you have a contact problem brewing. Same edge number, completely different situation.
GPU overheating symptoms: how to check if a GPU is overheating
The symptom list is your fastest diagnostic. Watch for these, in rough order of how often I see them:
- Performance that drops sharply after ten to twenty minutes of gaming, then recovers when you idle. That's throttling.
- Fans that ramp to full and stay there.
- Visual artifacts — flickering textures, colored speckles, lines that shouldn't be there.
- Sudden black screens mid-game.
- Full system shutdowns, not just the game crashing. This is the card or the PSU protecting itself.
The black screen and shutdown cases are the ones that send people to the forums. The honest answer is that they overlap heavily with power-delivery problems, so confirm the temperatures before you blame heat. A card that shuts down at 65 °C isn't overheating, no matter how much it feels like it.
Can you fix an overheated GPU?
Yes, and in most cases without buying a replacement. Work through the sequence: clean, verify airflow, tune the fan curve, undervolt. If temperatures are still bad and your hotspot delta is wide, it's time for thermal paste and possibly new thermal pads. That last step is where people hesitate, and I understand why — you're opening a card that cost real money. But repasting is a forty-minute job with basic tools, and it resurrected a card I'd written off entirely.
The only case you probably can't fix is a card with physical damage — a cracked die, a dead fan motor you can't source a replacement for, a VRM that's given up. Even then, a replacement fan assembly is cheap and often the whole fix.
My honest verdict on the whole process
Start with measurement, not with a screwdriver. I've watched people disassemble a perfectly healthy card because they never checked the hotspot, and I've watched them miss a dried-out paste job because the numbers looked "fine." The four figures from Step 1 are worth more than any forum thread. Get those, and the rest of this becomes obvious.
And when you do repaste, do it with intent. A card that ran cool for years and suddenly doesn't is telling you something specific. Listen to the delta, not the headline number, and you'll fix it the first time.
