Is Your GPU Too Hot? What NVIDIA's Temperature Limits Actually Mean
Some links in this guide are affiliate links - we may earn a commission at no extra cost to you.
For a few hours on 26 February 2026, NVIDIA shipped a driver that stopped some graphics card fans from spinning. Version 595.59 went out with day-one support for Resident Evil Requiem, and by 11am Pacific the same day NVIDIA had pulled the download and posted a note telling anyone already running it to roll back. Its own release notes later described the defect in two lines: monitoring software was not detecting all fans on the card, and one or more fans were not spinning after the driver update.
The interesting part was not the bug, which NVIDIA fixed with a full WHQL driver six days later. It was watching thousands of people open a monitoring tool for the first time, see a number they had never looked at before, and have no idea whether it was fine. Plenty of them panicked at 0 RPM on an idle desktop, which is exactly what a healthy modern card is supposed to do.
So this is the reference for that question. What NVIDIA's published temperature limits actually are, what they actually mean, and how to tell a real thermal problem from a normal one.
The number everyone quotes is a throttle point, not a danger line
Every NVIDIA product page lists a "Maximum GPU Temperature." It gets repeated in forum threads as the temperature at which your card is in trouble. That is not what it is.
NVIDIA's own monitoring documentation defines three separate thresholds, and the published one is the lowest of them:
- Max Operating Temperature is the temperature at which, in NVIDIA's words, the GPU's software will optimize its clock for thermal conditions. The driver trims boost clocks to hold the line. You lose a little performance. Nothing is harmed.
- Slowdown Temperature sits above it and is a hardware response, described as reducing core clocks by a factor of two or more. This is the card protecting itself firmly rather than gently.
- Shutdown Temperature is the last one, where the GPU turns itself off.
A card that touches its max operating temperature under a heavy load is a card doing its job. The protections stack in your favor, and they run whether or not you are paying attention.
The actual limits, by card
There is no single NVIDIA temperature limit, which is the most common mistake in this topic. Ampere and Ada are each internally consistent, but Blackwell is not, and the spread inside the 50-series is wide enough to matter.
| Generation | Card | Max GPU temperature |
|---|---|---|
| Ampere (RTX 30) | 3060, 3060 Ti, 3070, 3070 Ti, 3080, 3080 Ti, 3090 | 93C |
| Ampere (RTX 30) | 3090 Ti | 92C |
| Ada (RTX 40) | 4060, 4060 Ti, 4070 family, 4080, 4080 SUPER, 4090 | 90C |
| Blackwell (RTX 50) | 5090 | 90C |
| Blackwell (RTX 50) | 5060 | 89C |
| Blackwell (RTX 50) | 5080, 5070 Ti | 88C |
| Blackwell (RTX 50) | 5060 Ti | 87C |
| Blackwell (RTX 50) | 5070 | 85C |
Two things stand out. The RTX 5070 has the lowest ceiling of any card here at 85C, so a 5070 owner applying the familiar 90C figure from the 40-series is working with five degrees of margin that does not exist. And the 5060 is rated higher than the 5060 Ti, which reads like a typo and is not.
Figures above are NVIDIA's published maximum operating temperatures from its own product specification pages, checked August 2026. Free to cite with a link.
For context on what normal looks like rather than what the ceiling is, published measurements of Founders Edition cards put the RTX 3080 at roughly 34C idle and 78C gaming, the RTX 4090 at 37C idle and 66C gaming, and the RTX 5090 at 50C idle and 77C gaming. Gaming in the 70s is ordinary. None of those cards is anywhere near its limit.
Zero RPM at idle is the system working
The single most common false alarm is a stopped fan on an idle machine. Every current GeForce card ships with a quiet mode that turns the fans off entirely when there is nothing to cool, and the thresholds are published. ASUS documents them for its RTX 30, 40 and 50-series cards: the fans start above roughly 55C or above about 100W of draw, and they stop again below about 50C, below about 50W, with the zero-decibel mode enabled.
That means the correct way to check your fans is not to look at the desktop. It is to load the card and watch what happens. Run a game or a stress test, let the temperature climb past 55C, and confirm the fans spin up. Fans that stay at zero while the temperature keeps rising under load is the symptom that matters, and it is precisely what the February driver bug produced.
The number RTX 50-series owners cannot see
There is a real blind spot worth knowing about, and it has nothing to do with any driver bug.
The GPU temperature you see in most tools is the average across the die. The hot spot, the hottest single point on the silicon, typically runs meaningfully higher. NVIDIA removed public access to the hot spot sensor on Blackwell cards, so most monitoring tools simply do not show it on a 50-series GPU.
The gap can be dramatic. NVIDIA's internal tooling has been reported reading a 107C hot spot on an RTX 5070 Ti at a moment when public tools were showing 67 to 68C for the same card. Later versions of HWMonitor restored the reading. If you own a 50-series card and you are troubleshooting genuinely odd behavior, be aware that the comfortable number on your screen may not be the highest number on the die.
Memory temperature is worth a similar caution. The figure that circulates for GDDR6X is around 110C, which traces back to Igor's Lab citing industry contacts rather than to a published Micron specification, and no public limit for the GDDR7 on 50-series cards exists at all. Treat both as rough guidance, not spec.
How to actually check
Tool choice matters more than it sounds, because the useful detail is per-fan.
GPU-Z is the right answer. Its Sensors tab lists each fan on its own row, plus separate temperature sensors. That per-fan breakdown is the whole point: the characteristic failure is one of two or three fans stopping while the others keep spinning, which any averaged or single-fan reading hides completely. It is also what let people identify the February bug at all, since the fans were disappearing from the list rather than reporting zero.
Two tools that will not answer this question: Windows Task Manager shows GPU temperature but has never reported fan speed, and the NVIDIA App overlay reports utilization, frame rate and latency rather than RPM. Both are fine for a quick temperature glance and useless for diagnosing a fan.
If the temperatures genuinely are too high, the causes are almost always physical and boring rather than exotic. Dust in the heatsink fins is the leading one and the easiest to fix. After that: case airflow with no intake path to the card, a GPU sagging into a fan cable, ambient room temperature in a hot month, or dried-out thermal paste on a card that is several years old. Our case finder filters on airflow and clearance if the enclosure turns out to be the constraint, and the PC case guide covers what actually moves air rather than what looks like it does.
The driver footnote
For completeness, since this is where the topic came from. The affected driver was 595.59, released and withdrawn on 26 February 2026. NVIDIA's instruction at the time was to roll back to 591.86 for anyone experiencing fan control problems. The fix arrived as 595.71 on 2 March, a fully certified driver rather than a hotfix, and a separate 595.76 hotfix two days later addressed an unrelated voltage issue that 595.71 had introduced.
The practical situation today is that essentially nobody is still exposed. 595.59 was downloadable for a matter of hours and has been unavailable since, and anyone who has updated even once since early March is clear. Safe versions are 591.86 and earlier, or 595.71 and everything after it, which by now is a long list.
Two things worth stating plainly, because the coverage at the time blurred them. NVIDIA never specified which cards were affected, so the widely repeated "RTX 30, 40 and 50 series" framing is press aggregation rather than a manufacturer statement, and at least one credible user report involved an RTX 2060. And no case of a graphics card actually damaged by this bug was ever confirmed. Every "this could kill your GPU" headline described a mechanism, not an observed failure. The thermal protections above were active the entire time, which is the reason a stopped fan became a performance problem rather than a funeral.
The short version
Your GPU is probably fine. Gaming in the 70s or low 80s is normal, the published maximum is a throttle point rather than a damage threshold, and stopped fans on an idle desktop are the card being quiet on purpose. The numbers worth remembering are your specific card's ceiling from the table above, and the fact that the 50-series spreads from 85C to 90C instead of sharing one figure.
Check under load, not at idle. Use a tool that shows each fan separately. And if the temperatures really are high, look for dust and airflow before you look for anything clever.
Temperature limits sourced from NVIDIA's published product specifications and its monitoring documentation, checked August 2026. Driver timeline sourced from NVIDIA's own release notes for versions 595.59, 595.71 and 595.76.
Found this useful? Pass it along:
Frequently Asked Questions
What temperature is too hot for a GPU?
Higher than most people think, and it depends on the card. NVIDIA publishes a maximum operating temperature per model: 93C across most RTX 30-series cards, 90C for the RTX 40-series, and between 85C and 90C for the RTX 50-series depending on the model. Reaching that number is not damage. It is the point where the driver starts trimming clocks to hold the temperature, and there are two further thresholds above it before the card protects itself by shutting down.
Is 80C safe for a GPU?
Yes. 80C under a gaming load is normal and within spec for every current NVIDIA card. Measured results put an RTX 3080 Founders Edition around 78C while gaming and an RTX 5090 Founders Edition around 77C. If your card sits in the high 70s or low 80s under load, nothing is wrong.
Why are my GPU fans not spinning?
Most likely because the card is idle and that is the designed behavior. Modern GeForce cards stop their fans entirely below roughly 50C and 50W when the quiet fan mode is on, and start them again above about 55C or 100W. Zero RPM on the desktop is normal. Fans that stay stopped while the temperature climbs under load are the actual problem.
What is the max temperature for an RTX 5070?
85C, which is the lowest ceiling of any current NVIDIA card and lower than the 90C limit people often assume from the 40-series. The RTX 5080 and 5070 Ti are rated to 88C, the 5060 Ti to 87C, the 5060 to 89C, and the 5090 to 90C. Applying one number across the 50-series gets several models wrong.
How do I check if my GPU fans are working?
Use GPU-Z and look at the Sensors tab, which lists each fan separately. That per-fan view matters because the common failure is one fan stopping while the other keeps going, which a single averaged reading hides. Windows Task Manager shows GPU temperature but not fan speed, and the NVIDIA App overlay does not report fan RPM either.