GPU Hotspot Temp High but Core Temp Normal? What a Large Core/Hotspot Delta Means & Mounting-Pressure Diagnosis
Key Takeaways
Monitoring software shows two GPU temperatures: the average (edge) temperature and the hotspot (junction) temperature, which is the hottest sensor on the die. A large gap between them — typically above 20–25°C — usually points to poor cooler contact: uneven mounting pressure, displaced thermal paste, wrong pad thickness, or paste pump-out. This guide explains what the delta means, model-specific limits, and a step-by-step diagnosis that starts without opening the card.
Two Temperatures, One Die
Open HWiNFO or GPU-Z and you will often see two GPU temperature readings that can differ dramatically. Understanding the difference is the key to the whole topic.
GPU Temperature (edge / average). This is the reading from the sensor near the GPU die's surface or edge — effectively an average or edge reading. Importantly, this is also the sensor that drives fan control. The fan curve responds to this number, not to the hotspot.
Hotspot (AMD: Junction Temperature). Modern GPU dies carry many temperature sensors scattered across the chip. The hotspot is simply the highest reading among all of them at any moment. Under different workloads, the hottest point wanders — one moment it is near one shader cluster, the next near a memory controller — and the hotspot value follows whichever sensor is currently hottest.
This split has a real consequence. If the hotspot is far hotter than the average, the fans will not speed up in response, because the fans do not see the hotspot. The result: a die that is locally hot enough to throttle while the fans are still idling. A widely cited real-world example from Brazilian repair technician Paulo Gomes (reported via Club386) showed an RTX 5070 Ti averaging 68°C while its hotspot hit 107°C — already thermally throttling, with no fan response to speak of.
A short history of hotspot readings
AMD introduced the reading broadly with the Radeon VII (Q1 2019), then called T-junction; from the RX 5000 series on it was named Hot Spot / Junction. NVIDIA followed with hotspot readings on the RTX 30 series. On the RTX 50 series, NVIDIA hid hotspot from its official NvAPI — third-party tools like GPU-Z and HWiNFO initially could not read it. In 2026, modders re-exposed it from lower-level hardware data, and HWiNFO's newer versions followed. If you own an RTX 50 card and your older software shows no hotspot, that does not mean the problem cannot exist — it means NVIDIA hid the reading.
Why the Gap Exists
A GPU die does not heat evenly. Shader arrays, memory controllers and other blocks concentrate heat in specific spots, while the edge sensor sits somewhere cooler. How large the gap (delta) is, however, is largely determined by cooler contact quality:
- Uneven mounting pressure or screws tightened inconsistently
- Incomplete thermal paste coverage
- Paste pump-out — repeated heating and cooling cycles squeezing paste out from the die edges until the centre is left dry
- Thermal pads of the wrong thickness, jacking the heatsink up off the die
- PCB flex or GPU sag in the slot
Pump-out deserves special attention because it worsens over time. Igor's Lab documented an RTX 5090 teardown where the paste was completely displaced and the die centre was bone dry — the hotspot's smoking gun. One Overclockers forum user tracked the same benchmark at the same power draw over a week: core temperature rose 3.2°C, hotspot rose 9°C, and the delta widened from 9.3°C to 15.1°C. On disassembly, the paste had been squeezed out with a dry ring in the centre — the classic signature of pump-out in progress.
There is also a useful cross-check with memory temperature: the memory modules and power stages share the same heatsink. If memory temperatures have fallen while the hotspot has risen, the heat is likely trapped in the die and not reaching the heatsink at all — the heatsink has spare capacity for the memory precisely because it is not absorbing the die's heat. That pattern strongly suggests a contact problem.
Reading the Numbers: Deltas and Limits
There is no single official standard covering all models, but the community and official specs converge on a workable framework. Note that these are empirical rules of thumb, not iron laws — model variation matters.
| Delta (hotspot − core average) | Typical interpretation |
|---|---|
| ~5–10°C | Usually normal; good cooler contact |
| ~10–20°C | Mostly acceptable; watch if it widens over time |
| ~20–25°C+ | Usually points to a contact, mounting or airflow problem, especially if persistent |
AMD's official specification puts the junction ceiling at 110°C for the RX 5000, 6000 and 7000 series. BIOS data for RDNA4 (RX 9070 XT) interpreted by Igor's Lab shows limits of 110°C for both edge and hotspot, and 108°C for memory. AMD publicly stated that an RX 5700 running at 110°C junction was within specification, citing precise multi-sensor control as the reason.
Absolute values matter too: a hotspot sustained at 95–100°C or above on NVIDIA cards, or approaching 110°C on AMD, is warning territory regardless of what the delta says. Also, some AMD models are reported to ship from the factory with deltas above 25°C — so "20°C+ means broken" is a rule of thumb, not a law. Always combine the delta with the trend (is it widening?) and behaviour (is anything throttling?).
That the delta is not just a number was proven by a TechPowerUp case study on a Gigabyte RTX 2080 with open benchmark data: before reworking the thermal interface, the hotspot sat at 107.2°C with a 28.1°C delta, fans at 77.9%, and the clock pinned at 1837 MHz. After the rework: hotspot 78.3°C, delta 12.5°C, fans at 38.5%, clock at 1924 MHz, and benchmark score up 25%. A large delta converts directly into lost performance and lost silence.
Step-by-Step Diagnosis: Contact Problem or Normal?
Preparation. Use HWiNFO64 (sensors-only) or GPU-Z. Run a repeatable load — the same game scene, FurMark or 3DMark — and log core temperature, hotspot, memory temperature, fan speed and clocks.
Step 1 — Look at absolute values first. Hotspot sustained at or above ~95–100°C (NVIDIA) or near 110°C (AMD) means action is needed regardless of the delta.
Step 2 — Look at the delta. Persistently above 20–25°C suggests a contact issue; 10–20°C means observe first.
Step 3 — Look at the trend. Compare with last week or last month. A widening delta means pump-out or paste drying is in progress — one forum case widened by about 6°C in a week.
Step 4 — Rule out airflow. Test again with the side panel removed. If the hotspot drops significantly, case airflow is poor. If nothing changes, the problem is not airflow — it points to contact.
Step 5 — Look for behavioural evidence. Clocks failing to hold, benchmark scores dropping, fans at high speed for no apparent reason — the TechPowerUp case shows a large delta turns into real performance loss.
Step 6 — Cross-check with memory temperature. Memory normal-to-cool but hotspot high? Heat is trapped in the die and not reaching the heatsink — contact is the likely culprit.
Step 7 — Rule out software causes. Did the delta appear suddenly after a driver update or a power-limit change? That suggests changed power behaviour. A slow, gradual worsening points to the hardware or the interface material.
Step 8 — Last resort: open the card and inspect the paste. The heatsink-off paste imprint should be thin and even; a dry centre with no paste, or thick on one side and thin on the other, indicates pump-out or uneven pressure. Caveat from Igor's Lab: an uneven imprint does not automatically prove bad factory application — the disassembly process itself disturbs the distribution, so interpret with care. Warning: opening the card may void the warranty.
Things you can do without opening the card (diagnosis doubles as treatment): clean dust from the cooler, adjust the fan curve, improve case airflow, lower the power limit or undervolt, check for GPU sag that could be skewing mounting pressure, and reseat the PCIe screws so the card sits straight.
When It Is Normal and When to Worry
Normal
- Delta of 5–15°C, hotspot well away from the ceiling, clocks and benchmarks normal.
- High-power AMD cards or aggressively designed coolers can acceptably run junctions of 85–95°C in games (model-dependent).
- Brief spikes (delta jumping for a moment) may just be a power transient the fan curve has not caught up with — not necessarily a contact problem.
- RTX 50 owners seeing no hotspot reading: normal (NVIDIA hid it); newer HWiNFO versions or HWMonitor 1.65+ can read it.
Warning signs
- Hotspot sustained at 100°C+ (NVIDIA) / 105°C+ (AMD), or delta persistently at 25°C+.
- Delta widening over time.
- Throttling, falling benchmark scores, fans permanently at full speed.
- A new card with a sustained delta of ~40°C — one forum case showed the owner's card at 40°C while a friend's identical card sat at 22–25°C; that is clearly abnormal.
Common Misconceptions
"Core at 60-something means the card is cool." A cool average says nothing about the hottest point. The Paulo Gomes case: 68°C average, 107°C hotspot, already throttling.
"Hotspot near 110°C means it is broken." AMD officially considers up to 110°C within spec (RX 5700 statement). But "within spec" is not "ideal" — sustained operation at the ceiling costs you fan noise and boost clocks.
"No hotspot reading means no hotspot problem." On RTX 50 cards it is hidden by NVIDIA, not absent.
"A big delta always means the paste is dried out." It could also be mounting pressure, pad thickness, PCB flex or airflow — and Igor's Lab warns that paste-imprint reading has its traps.
"Expensive paste will definitely shrink the delta." Paste that is too runny gets pumped out; some pastes suit CPUs but not GPUs (forum reports: watery pastes migrate away through thermal cycling). Thick pastes or PTM7950 phase-change pads are common community approaches, but that is community practice, not an established conclusion.
"A high hotspot means the card is badly designed." Large model-to-model delta variation mostly comes down to individual contact issues — before/after repaste comparisons (30°C → 20°C, 21°C → 12.5°C) show this is usually not a design flaw.
FAQ
Q1: My core temperature is 65°C but my hotspot is 100°C. Is my card dying?
Not necessarily dying, but that 35°C delta is in warning territory and needs investigation. Start with the no-disassembly steps: check the trend over time, test with the side panel off, cross-check memory temperature, and look for throttling or benchmark drops. If the delta persists or widens, the cooler-to-die contact is the prime suspect. Either way, 100°C sustained is hot enough to deserve action.
Q2: What is a normal hotspot delta?
Roughly 5–10°C is usually normal, 10–20°C mostly acceptable, and 20–25°C+ typically points to a contact, mounting or airflow problem — but these are empirical rules of thumb, not iron laws. Some models ship with larger deltas from the factory, so always combine the number with the trend (widening?) and behaviour (throttling? noise?).
Q3: Do I need to repaste if my delta is 22°C?
Not immediately. A 22°C delta sits in the "observe" band. First rule out airflow (side-panel test), check whether it is widening over weeks, and confirm nothing is throttling. If it keeps widening or the hotspot approaches its ceiling, then a thermal-interface rework — paste or a phase-change pad, done carefully, noting warranty implications — is the logical next step. Also check simpler causes first: dust, fan curve, GPU sag, loose mounting screws.
Q4: Why can't I see a hotspot reading on my RTX 50 series card?
NVIDIA hid the hotspot reading from its official NvAPI on the RTX 50 series, so older monitoring software shows nothing — that is expected, not a sign your card is fine or broken. Newer HWiNFO versions (following modders who re-exposed the data from lower-level hardware registers in 2026) and HWMonitor 1.65+ can display it.
Q5: Will repasting void my warranty?
In many cases, yes — opening the card to replace thermal paste or pads can void the manufacturer's warranty, and policies vary by brand and region. That is why the diagnostic steps above deliberately start with non-invasive tests (airflow, trends, fan curves, sag checks). If the card is still under warranty and the delta is clearly abnormal, contact the manufacturer or your seller first before opening anything.
Q6: What is pump-out, and how do I prevent it?
Pump-out is the gradual displacement of thermal paste from between the die and the heatsink as repeated heating/cooling cycles squeeze it out past the die edges, leaving the centre dry — often with a dry ring pattern on disassembly. It shows up as a delta that widens over weeks or months. Community practice favours thicker pastes or phase-change pads (like PTM7950) as more resistant, but this is community experience rather than an established conclusion. Varies by model and paste chemistry.