GPU Boost Clock and Temperature Throttling: How Thermal Limits Compress Your Clock Speed
GPUs do not run at a fixed clock speed. The boost algorithm constantly adjusts frequency based on temperature, power draw, voltage, and current. Temperature alone can compress sustained clock speed by 10 to 15 percent in some configurations—without any indication in a standard benchmark number.
The maximum boost clock advertised for any GPU is a ceiling, not a sustained operating point. It represents the highest frequency the card will run under favorable conditions: cool temperatures, adequate power delivery, and a workload that does not saturate the power limit. Under sustained gaming loads, the effective clock speed that the GPU actually maintains over a 30-minute session is almost always lower than the boost specification.
NVIDIA's Boost 4.0 Temperature Curve
NVIDIA's GPU Boost algorithm, in its current implementation, has two relevant temperature thresholds. The first is the Temperature Target (T-Target), which is set in the VBIOS and typically ranges from 80 to 90 degrees Celsius on desktop Ada Lovelace and Ampere cards. Below this temperature, the boost algorithm operates normally and reaches its maximum frequency given the power budget. Above T-Target, the algorithm begins applying a negative clock offset, scaling the frequency down as temperature climbs further.
The second threshold is the thermal shutdown point, which is typically 110 degrees Celsius for the GPU junction temperature on desktop cards. Between T-Target and the shutdown point, the card is throttling based on temperature. This is visible in HWiNFO64 as a "Temp Limit" throttle flag alongside the actual clock reading.
In practice, most well-cooled desktop cards stay below T-Target under gaming loads and do not experience temperature-based throttling. The scenario where it matters is a compact card in a restricted airflow case, or a laptop GPU pushing its rated TGP limit with poor chassis ventilation.
NVIDIA Temperature vs Sustained Clock Example (RTX 4070)
| GPU Temp (GPU Core) | Approximate Sustained Clock | Throttle Active? | Performance vs Baseline |
|---|---|---|---|
| 60–75 °C | ~2,550 MHz | No | 100% |
| 76–83 °C | ~2,520–2,550 MHz | No | ~99–100% |
| 84–87 °C (T-Target) | ~2,460–2,510 MHz | Yes (minor) | ~96–98% |
| 88–93 °C | ~2,340–2,430 MHz | Yes (moderate) | ~91–95% |
| 94+ °C | <2,300 MHz | Yes (hard) | <90% |
AMD's Approach: Power First, Temperature Second
AMD RDNA 3 GPUs use a different prioritization. The primary clock governor is the power limit (Socket Power), not temperature. The card maximizes frequency within the available power budget, and temperature reduction is handled primarily through the fan curve rather than frequency scaling. If the card hits its thermal limit (typically 110 degrees Celsius junction on desktop cards), frequency does drop, but in a well-cooled system this should not happen under gaming loads.
The implication for AMD GPU owners is that aggressive power limit tuning has a larger immediate impact on performance than thermal management, unless the card is genuinely running hot. Overclocking by raising the power limit first, then checking temperatures, is the correct sequencing. Temperature-based throttling on RDNA 3 under gaming conditions in a decent case is uncommon at stock settings.
Raising the T-Target on NVIDIA GPUs
MSI Afterburner exposes a Temperature Limit slider that adjusts the T-Target on most NVIDIA cards. Moving it from 83 to 87 degrees Celsius gives the boost algorithm more headroom before it begins frequency reduction. The card will run hotter and the fans will spin faster, but sustained clock speed under prolonged loads improves measurably.
The default T-Target is a noise and longevity compromise. Raising it by 3 to 5 degrees does not materially affect GPU lifespan at the temperatures involved but does allow the fan curve to reach higher RPM before the boost algorithm starts reducing clocks. Whether the additional fan noise is acceptable is the only real constraint on this adjustment.
Practical Takeaways
- Monitor GPU core clock over a 30-minute gaming session, not just the peak boost spec, to see the actual sustained frequency your card achieves.
- If HWiNFO64 shows a "Temp Limit" flag concurrent with clock drops, thermal management is actively compressing performance.
- Raising the T-Target by 3–5 degrees in Afterburner is often a zero-cost fix for temperature-limited NVIDIA cards that run warm in restrictive cases.
- For AMD cards, check whether power limit is the active constraint before addressing thermal conditions.
- On laptops, the thermal throttle thresholds are set by the OEM VBIOS and are less user-adjustable without VBIOS modification.