Edge and hotspot: what each sensor reads
A modern GPU has several temperature sensors. Three matter most:
- Edge (main GPU temperature): the average or primary reading of the die. Most tools and spec sheets quote this number.
- Hotspot (junction): the highest reading across many sensors on the die. It is a real, localised temperature, not a fake or a miscalibrated reading.
- Memory junction: a separate sensor for the GDDR memory chips.
You can read the hotspot in HWiNFO64 under “GPU Hot Spot Temperature”, or in GPU-Z.
What a normal gap looks like
Under load, the hotspot is usually 10–25°C above the edge reading. A gap of 15–20°C is normal and is not a defect.
The gap depends on the card, its cooler, its power draw and the workload. There is no single correct number, so judge the gap against the same card’s behaviour over time rather than against a number from another model.
When the gap is a warning sign
A gap above about 30°C suggests one of these:
- An uneven mount, where the cooler presses harder on one side of the die
- Paste that has pumped out or dried up
- A thermal pad that is too thick or too thin
Watch the trend. A gap that grows over months, while the edge temperature stays about the same, points to the cooler’s contact with the die. A hotspot that reaches the card’s limit, or clocks that drop while the fans run flat out, also means the cooling needs attention.
How to check the gap properly
- Use the same test every time. Pick one game scene or benchmark, and run it for about 20 minutes.
- Restore stock settings first. Undo any overclock, undervolt, power-limit change or custom fan curve before you measure.
- Log the key numbers. Record the edge, the hotspot, the fan speed, the board power and the room temperature.
- Compare at the same moment. Read both sensors after the card has settled under a steady load. Idle readings are less useful, because the fans may behave differently.
- Ignore single peaks. A short spike during a loading screen tells you little.
A warmer room raises both readings, so note the room temperature with each log. If both rise together, the case, the room or the fans are the likely cause. If the edge stays similar and only the gap grows, the cooler contact is the stronger suspect.
Limits differ: judge each sensor against its own limit
- AMD RX 6000, 7000 and 9000: the hotspot (junction) limit is 110°C. Up to about 100°C in games is normal.
- NVIDIA RTX 30 and 40: the maximum edge temperature is around 90–93°C. The hotspot is generally fine below about 100–105°C.
- Newer cards: some RTX 50 series models do not show a hotspot reading in most tools, as reported in 2025. Check what your monitoring tool supports.
Each sensor has its own limit, so judge the hotspot against the hotspot limit and the edge against the edge limit.
Memory junction is a third number
The GDDR memory junction is commonly 90–105°C under load, and it throttles near 110°C on many cards. A hot memory reading is not the same as a high hotspot. The core gap does not show whether the memory pads are making good contact, so check the memory reading as well if your card exposes it.
Before you repaste
Repasting is not the first fix, and it is not always allowed.
- Warranty: repasting or re-padding may void the warranty, depending on the brand and region. Check before you open the card.
- Pads: thermal pad thickness is model-specific. A pad that is too thick can lift the cooler away from the die and make the hotspot worse.
- Mounting: tightening cooler screws by feel is not a fix. Mounting pressure is engineered, so follow the maker’s guidance.
- Materials and safety: repaste only with evidence, the right materials and the skill to do it. Power off, unplug, and discharge static first, because thermal pads tear easily.
For the other readings to check first, see what your GPU temperature should be while gaming. To check whether heat is slowing the card, see thermal throttling.