Reference
GPU hotspot delta and when to repaste
AMD GPUs on Linux expose two temperature readings, not one: edge and hotspot (junction). The gap between them is one of the earliest, quietest signs that thermal paste is drying out, weeks before it shows up as a crash or a throttling problem you would actually notice while playing.
Edge temperature vs hotspot (junction) temperature
Edge temperature is an average across the whole GPU die, it is the number most monitoring tools show as "GPU temperature." Hotspot, also called junction temperature, is a reading from the single hottest point on the die, typically right over the core. Every GPU has a gap between the two, a gap of 10 to 15°C under full load is normal on plenty of consumer cards straight from the factory.
Why the gap grows
Thermal paste between the die and the vapor chamber or heatsink dries out over months of thermal cycling, heating up and cooling down thousands of times. As contact efficiency degrades, the hottest point on the die climbs faster than the broader edge average, so the delta between edge and hotspot widens even though your case airflow, ambient temperature, and fan curve have not changed at all. This is why the delta is a better early signal than either temperature alone.
The actual threshold
| Signal | What it means |
|---|---|
| 10 to 15°C delta under load | Normal, no action needed |
| Delta grows 6°C or more over a 60-day window | Repaste recommended, drying thermal paste |
| Example: delta grows from 8°C to 19°C over 60 days | An 11°C increase, well past the threshold, clear repaste signal |
The 60-day window matters as much as the threshold itself, thermal paste degradation is a slow trend, not a single bad reading. A one-off high delta on a hot day tells you nothing, a delta that keeps climbing measured a day apart, over two months, is the real signal.
Why this is AMD-only, for now
AMD GPUs expose both edge and junction temperature through the standard Linuxamdgpu driver. NVIDIA's NVML library, on Linux, only exposes die (edge) temperature, not junction, so there is currently no public way to read this signal for NVIDIA cards on Linux at all, this is a driver and API limitation, not a choice.
How Thermal Canary Pro tracks this
The Hotspot Delta Analyzer records an edge-to-hotspot sample roughly once a day in the background, keeps a rolling 60-day history, and surfaces a plain suggestion in the Health panel the moment the delta growth crosses the threshold above. You do not have to remember to check it or do the subtraction yourself.
FAQ
What is the difference between GPU edge temperature and hotspot or junction temperature?
Edge temperature is an average reading across the whole GPU die. Hotspot, also called junction temperature, is a reading from the single hottest point on the die, usually directly over the core. The gap between them, the delta, exists on every GPU and is not itself a problem.
What is a normal edge-to-hotspot delta?
A gap of 10 to 15°C under full load is routine on plenty of consumer cards straight from the factory. What matters is not the size of the gap at any one moment, it is whether that gap grows over weeks of normal use.
Why does the hotspot delta grow over time?
Thermal paste between the GPU die and the vapor chamber or heatsink dries out and loses contact efficiency over months of thermal cycling. As it degrades, the hottest point on the die (junction) climbs faster than the broader edge average, so the gap between the two widens even though nothing else about your cooling changed.
How much delta growth actually means I should repaste?
Thermal Canary Pro's Hotspot Delta Analyzer flags it once the edge-to-hotspot delta has grown by 6°C or more over a 60-day trend window. As a concrete example: a delta that grew from 8°C to 19°C over 60 days, an 11°C increase, is a clear repaste signal, well past that threshold.
Does this apply to NVIDIA GPUs too?
Not on Linux today. NVIDIA GPU junction temperature tracking depends on NVML exposing that data, and NVML on Linux only exposes die (edge) temperature, not junction. This is an AMD-only capability for now, since AMD GPUs expose both edge and junction temperature through the standard Linux amdgpu driver.
Do I need to track this manually?
You can, by logging edge and junction temperature yourself over time and watching the gap. Thermal Canary Pro's Hotspot Delta Analyzer does this automatically in the background, roughly once a day, and surfaces a plain-English suggestion in the Health panel once the trend crosses the threshold, so it is not something you have to remember to check.
General guidance based on typical consumer AMD GPU behavior on Linux. Repasting a GPU voids some manufacturer warranties, check your card's warranty terms before opening it.