A safe GPU temperature is a core reading below your card’s published maximum, with no thermal slowdown. NVIDIA lists 90 °C for every desktop RTX 40 card and 85 to 90 °C for RTX 50, from 85 °C on the RTX 5070 to 90 °C on the RTX 5090. AMD’s RX 9000 and Intel’s Arc B580 spec pages list none.
At a glance
- RTX 40 desktop
- 90 °C maximum on every model
- RTX 50 desktop
- 85 °C (RTX 5070) to 90 °C (RTX 5090)
- Radeon and Arc
- No maximum on the RX 9000 or Arc B580 spec pages
- Read your own
- nvidia-smi: temperature.gpu plus temperature.gpu.tlimit
Provenance
- Oct 4, 2026Published
- Nov 18, 2026Last substantive update
A safe GPU temperature is a core reading below your card’s published maximum, with no thermal slowdown. NVIDIA publishes that maximum for its desktop RTX 40 and RTX 50 cards: 90 °C across the RTX 40 series and 85 to 90 °C across the RTX 50 series. AMD and Intel don’t list one on the RX 9000 and Arc B580 specification pages. On a supported NVIDIA card you can also read the limit straight off the card with nvidia-smi.
Maximum GPU temperature by card
These are NVIDIA’s Maximum GPU Temperature figures for its desktop Founders Edition or reference designs, read on its specification pages on November 18, 2026. NVIDIA says add-in-card specifications may vary, so use the board maker’s page for a custom card when it publishes a different limit.
| GPU | NVIDIA maximum | In Fahrenheit |
|---|---|---|
| GeForce RTX 5090 | 90 °C | 194 °F |
| GeForce RTX 5080 | 88 °C | 190.4 °F |
| GeForce RTX 5070 Ti | 88 °C | 190.4 °F |
| GeForce RTX 5070 | 85 °C | 185 °F |
| GeForce RTX 5060 Ti | 87 °C | 188.6 °F |
| GeForce RTX 5060 | 89 °C | 192.2 °F |
| GeForce RTX 5050 | 87 °C | 188.6 °F |
| GeForce RTX 4090 | 90 °C | 194 °F |
| GeForce RTX 4080 SUPER, RTX 4080 | 90 °C | 194 °F |
| GeForce RTX 4070 Ti SUPER, 4070 Ti, 4070 SUPER, 4070 | 90 °C | 194 °F |
| GeForce RTX 4060 Ti, RTX 4060 | 90 °C | 194 °F |
Every RTX 40 card shares 90 °C. The RTX 50 figures spread across five degrees and don’t follow the model order: NVIDIA lists 89 °C for the RTX 5060 and 85 °C for the RTX 5070. An RTX 5070 at 85 °C is already at its maximum, while an RTX 5090 or any RTX 40 card at the same reading still has 5 °C left. That spread is why copied temperature charts fail.
Radeon and Arc
AMD’s specification pages for the RX 9070 XT, RX 9070 and RX 9060 XT list board power and clocks but no maximum temperature, and Intel’s Arc B580 specification page has no thermal field at all.
AMD did publish a junction figure in 2019, for the RX 5700 series. In its Radeon community update, AMD explained that those cards keep raising clocks until any one of their many sensors reaches a 110 °C junction, or hotspot, temperature, and called running at up to 110 °C junction during typical gaming expected and within spec. That was said about the RX 5700 series. Don’t carry it to an RDNA 4 card; check your board maker’s page and watch whether clocks fall under load.
What “normal” means at idle and while gaming
Start with the model.
Normal is a relationship between the exact card, its workload and its own limit. An idle reading should be compared with that card’s idle baseline. A gaming reading should be captured during the game you actually play, then checked for remaining headroom and falling clocks. The manufacturer sources reviewed here do not prescribe one universal idle, gaming or “optimal” temperature range.
A PC also has no single “computer temperature.” CPU, GPU core, GPU hotspot, memory and storage sensors have different limits. Use this page for the GPU core; use the CPU temperature guide for the processor, and do not judge one sensor by another component’s range.
For Fahrenheit, use (C × 9/5) + 32. NVIDIA’s 85 C desktop RTX 5070 maximum converts to 185 F, while the 90 C maximum of the RTX 5090 and every RTX 40 card converts to 194 F. Those converted values are limits, not recommended targets. If you need the live reading first, use the Windows and command-line temperature checks.
Read the exact limit with nvidia-smi
Open Terminal or PowerShell on a PC with an NVIDIA driver and run:
nvidia-smi --query-gpu=name,temperature.gpu,temperature.gpu.tlimit --format=csv
The two temperature fields have different jobs:
temperature.gpuis the current core temperature.temperature.gpu.tlimitis the number of degrees remaining before the maximum operating temperature.
Add them.
current core temperature + current T.Limit = maximum operating temperature
That is the whole calculation.
NVIDIA’s current nvidia-smi documentation defines T.Limit as a signed, relative reading rather than an absolute temperature and warns that the sensor is unavailable on some products; if the command returns N/A, skip to the specification-page method below.
The RTX 5080 result on this site
On September 14, 2026, my RTX 5080 on driver 616.56 returned 42 C for the core and 46 C for current T.Limit:
42 C + 46 C = 88 C
A later reading returned 41 C plus 47 C, again 88 C, matching the Maximum GPU Temperature on NVIDIA’s RTX 5080 specification page.
The method matters more than the idle reading. Although the current temperature moves with the workload, the sum identifies the ceiling built into the card’s current temperature-limit reporting, which is the number a generic range cannot supply.
What zero and negative T.Limit values mean
For the same RTX 5080, nvidia-smi -q -d TEMPERATURE reported this ladder:
| nvidia-smi field | Reported value | Temperature on this 88 C card | What NVIDIA says can happen |
|---|---|---|---|
| Current T.Limit | 46 C | Core was 42 C | 46 C of operating headroom remained |
| Max Operating T.Limit | 0 C | 88 C | Software may optimize clocks for thermal conditions |
| Slowdown T.Limit | -2 C | 90 C | Hardware may optimize clocks to cool the GPU |
| Shutdown T.Limit | -5 C specification | Above 93 C | Below this margin, the GPU may shut down for protection |
Stop at zero. On this card, 88 C is the maximum operating point. The negative slowdown and shutdown values describe protection beyond that point; 93 C belongs to the protection ladder rather than the usable operating range.
NVIDIA’s manual says a current T.Limit of 0 C or lower means the GPU may optimize its clock based on thermal conditions. That makes the reading easy to use:
- Positive T.Limit: operating headroom remains.
- Zero T.Limit: the core has reached maximum operating temperature.
- Negative T.Limit: the core has passed that maximum and is in the protection ladder.
Hotspot and memory readings
NVIDIA’s maximum and the T.Limit equation are for the GPU temperature, the core reading nvidia-smi reports as temperature.gpu. The same nvidia-smi documentation lists a separate GPU Current Max Temp, taken from the hottest part of the GPU, and a memory temperature, each only on supported devices. NVIDIA’s specification pages publish no maximum for either. So don’t hold a hotspot or memory reading against the core limit, and don’t treat a hotspot figure from another card or brand as a rule for yours.
On RTX 50 cards the hotspot reading came back to HWMonitor in version 1.65.1 on July 16, 2026, and version 1.66 improved hotspot reporting on Radeon RX 9000 cards, according to CPUID’s version history. If the core stays below its maximum but the hotspot runs much warmer, the GPU hotspot guide and the GPU overheating guide explain the core-to-hotspot gap and what a widening gap changes about the diagnosis.
How much headroom should a healthy card have?
The NVIDIA sources reviewed here do not prescribe one fixed margin for daily use.
Start with the card. Then look at the workload and its behavior.
My hand-tuned ROG Astral RTX 5080 produced a useful real load check: across two recording windows of about 16 and 10 minutes, the core averaged 56 C and the highest of 308 readings was 60 C, 28 C below its published maximum. The throttle register read zero on every recorded row. The full log and method are in the ROG Astral RTX 5080 review and its open telemetry dataset.
That is one card, one cooler and one case. The logger paused for nearly an hour between its two recording windows. Use 60 C as the highest temperature it sampled; the file has no readings during the pause.
That log is from the card’s hand-tuned profile, whose fans run louder (up to about 51 dBA, against near-silent at stock, in the Astral bench dataset). On its stock profile the same dataset records 63 to 72 °C while gaming, 16 to 25 °C below the 88 °C maximum. Read the 60 C peak as the louder profile’s result, not the stock one.
Look at three things under the workload that matters:
- Headroom: a positive T.Limit means the core remains below its maximum.
- Clock behavior: a temperature at the limit with falling clocks means cooling is controlling performance.
- Stability: crashes, shutdowns or a temperature that suddenly rose from the card’s own baseline need diagnosis even when a generic chart calls the number normal.
If the card runs at its limit
If the core reaches its maximum during ordinary gaming, start with the fans: the GPU fans not spinning and GPU fans stuck at 100 percent guides cover the two common fan faults. The GPU overheating guide separates airflow, dust and cooler-contact problems, and the thermal throttling fix covers clocks that fall under heat. That page owns the fault diagnosis; this one owns the limit.
To run cooler without changing hardware, an undervolt lowers the voltage at the clock you choose. The RTX 5080, RTX 5070 Ti, RTX 5090 and RX 9070 XT guides each start from a sourced setting and a stability test.
If T.Limit is unavailable
Run the temperature query anyway. temperature.gpu works on NVIDIA discrete GPUs covered by the manual even when the relative limit sensor does not.
Do not guess. Find the exact model on the GPU maker’s specification page and look for Maximum GPU Temperature, Maximum Operating Temperature or a model-specific thermal limit. NVIDIA’s own RTX 50 table shows that adjacent models can differ.
For an add-in card, check its board maker; for a laptop, check the laptop maker, because NVIDIA’s desktop specification pages describe reference or Founders Edition designs and explicitly say board-partner specifications may vary rather than defining every card sold under the same GPU name.
If you only need the current number, the GPU section in the temperature-check guide covers Task Manager and the shorter nvidia-smi command.
Frequently asked questions
What is a safe GPU temperature?
Is 80 C safe for a GPU?
Is 85 C safe for a GPU?
Is 90 C bad for a GPU?
What is the maximum temperature for an RX 9070 XT?
What does T.Limit mean in nvidia-smi?
How do I find my GPU’s maximum temperature?
Is GPU hotspot temperature the same as GPU temperature?
What is a normal GPU temperature in Fahrenheit?
Sources and corrections
- Last updated
- Sources checked
- Methodology
- See our methodology for research and review standards. It draws on 16 cited sources, listed below, each checked against the original page on the date above. On September 14, 2026, I read the temperature fields from my NVIDIA GeForce RTX 5080 with nvidia-smi 616.56. One reading was 42 C with 46 C of current T.Limit margin; a later reading was 41 C with 47 C of margin. Both sums equal NVIDIA’s published 88 C maximum for the RTX 5080. A 90-second idle log contained 47 samples: 46 read 42 C plus 46 C, and the final row read 41 C plus 46 C. The separate values are integer readings, so I do not use the final 87 C sum as a precision claim. The same command reported Max Operating T.Limit 0 C, Slowdown T.Limit -2 C and Shutdown T.Limit -5 C. I did not capture T.Limit under load. The load example comes from the site’s existing June 28 first-party gaming log: 308 samples from the same RTX 5080, with a 56 C average and a highest recorded temperature of 60 C. Every sampled throttle flag was zero; there are no readings from the pause between the two logging windows.
- Sources
- NVIDIA: nvidia-smi documentation, temperature fields (T.Limit, GPU Current Max Temp, memory temperature) accessed 2026-11-18
- NVIDIA: GeForce RTX 5090 specifications (Maximum GPU Temperature 90 C) accessed 2026-11-18
- NVIDIA: GeForce RTX 5080 specifications (88 C) accessed 2026-11-18
- NVIDIA: GeForce RTX 5070 family specifications (RTX 5070 Ti 88 C, RTX 5070 85 C) accessed 2026-11-18
- NVIDIA: GeForce RTX 5060 family specifications (RTX 5060 Ti 87 C, RTX 5060 89 C) accessed 2026-11-18
- NVIDIA: GeForce RTX 5050 specifications (87 C) accessed 2026-11-18
- NVIDIA: GeForce RTX 4090 specifications (90 C) accessed 2026-11-18
- NVIDIA: GeForce RTX 4080 family specifications (90 C) accessed 2026-11-18
- NVIDIA: GeForce RTX 4070 family specifications (90 C) accessed 2026-11-18
- NVIDIA: GeForce RTX 4060 family specifications (90 C) accessed 2026-11-18
- AMD: Radeon RX 9070 XT specifications (no temperature limit listed) accessed 2026-11-18
- AMD: Radeon RX 9070 specifications (no temperature limit listed) accessed 2026-11-18
- AMD: Radeon RX 9060 XT specifications (no temperature limit listed) accessed 2026-11-18
- AMD Community: AMD Radeon Community Update, RX 5700 series junction temperature (August 2019, archived January 27, 2025) accessed 2026-11-18
- Intel: Arc B580 Graphics specifications (no thermal field) accessed 2026-11-18
- CPUID: HWMonitor version history (1.65.1 RTX 50 hotspot fix, 1.66 RX 9000 hotspot) accessed 2026-11-18
- Update log
- 2026-11-18 — Added NVIDIA’s limits for every desktop RTX 40 card, what AMD and Intel publish for Radeon and Arc, a hotspot and memory section, and FAQ answers that give the number; every source re-read.
- Corrections
- Spotted an error? Email a correction with the affected line and what you found. The draft includes this page’s address. Verified changes are recorded here. Contact details.