- Top Pick: Noctua NT-H1
- Runner-up: Noctua NT-H2
- Also consider: Thermal Grizzly Kryonaut
Open any hardware monitor during a game and a modern graphics card reports two temperatures that refuse to agree: the GPU temperature and the GPU hotspot. One might sit at 68 degrees Celsius while the other touches 84, and the gap is wide enough to make anyone wonder which number is lying. Neither is. They measure different things, and the distance between them, called the delta, is one of the most useful health signals a graphics card produces. This guide explains what the hotspot sensor measures, why it runs hotter, what gap is normal, and what it means when that gap grows.
As an Amazon Associate, we earn from qualifying purchases.
Quick Answer
The GPU hotspot is the single hottest point on the graphics chip, reported by the hottest of many on-die sensors, while the regular GPU temperature is an average or edge reading. A hotspot running 10 to 20 degrees Celsius above the core under load is normal. A gap that keeps growing usually signals aging thermal paste or uneven cooler contact.
Two Numbers, One Chip
The reading labeled GPU temperature is the conservative one. Depending on the card, it is either an average taken across the die or a single sensor placed near the edge of the chip, and it is the number older tools have displayed for decades. It moves slowly because it smooths over what is happening in the middle of the silicon.
The hotspot is the opposite philosophy. Modern graphics processors carry a network of temperature sensors spread across the die, and the hotspot value is simply the highest reading among them at that moment. AMD made this value prominent with its junction temperature reporting, and NVIDIA cards expose an equivalent hotspot reading through current monitoring tools. Because it tracks the single warmest spot rather than an average, the hotspot reacts within seconds when a demanding scene loads.
There is a third number worth knowing: memory temperature. Cards with GDDR6 or GDDR6X memory report a memory junction reading from the memory chips, not the processor die. It follows its own limits and cooling hardware, mainly thermal pads, so it should never be mixed into a core-versus-hotspot comparison.
Why the Hotspot Always Runs Hotter
Heat on a graphics die is not spread evenly. The shader arrays, cache, and control logic packed into the center of the chip work harder than the quieter regions near the edges. When a game pushes the card, those dense regions generate intense local heat in a very small area, while the edge sensor feeding the headline temperature sits in a calmer neighborhood.
Boost behavior widens the gap further. Modern cards raise clocks aggressively whenever there is thermal and power headroom, concentrating even more work, and therefore more heat, into the busiest parts of the die. The result is a hotspot that legitimately runs 10 to 20 degrees Celsius above the averaged reading under sustained load. That spread is not a defect or a sensor error; it is what the inside of a hard-working chip actually looks like. It is also why manufacturers set the throttling limits against the hotspot rather than the average, with the hotspot ceiling on many recent cards sitting around 105 to 110 degrees Celsius.
The Delta Is the Diagnostic
Because the absolute hotspot number depends heavily on the card model, cooler design, and workload, experienced builders watch the delta instead: hotspot minus core temperature, taken under the same load. On a healthy card with fresh paste and even mounting pressure, that gap usually lands between roughly 10 and 20 degrees Celsius in a demanding game, and it stays consistent over time.
The warning sign is change. A card that shipped with a 14-degree gap and now shows 25 or 30 degrees under the same game is telling a specific story. The usual cause is the thermal paste between the die and the cold plate. Over hundreds of heat cycles, paste can dry out or slowly migrate away from the center of the die, a behavior often called pump-out, leaving the hottest region with the thinnest coverage. Uneven mounting pressure after a cooler refit can produce the same signature. In each case the average temperature rises only slightly, because most of the die is still cooled well, while the hotspot climbs sharply.
A very wide delta at idle, by contrast, means almost nothing. At desktop loads the sensors sit close to ambient and gaps of a few degrees appear and vanish randomly. The delta only becomes a trustworthy signal under a sustained, repeatable load.
How to Read Your Own Card
Reading a delta properly takes about ten minutes and one monitoring tool. Software such as HWiNFO or GPU-Z exposes the hotspot and memory readings alongside the core temperature and can log them over time. The method matters more than the tool:
- Pick a repeatable load: the same game scene or benchmark loop every time.
- Let the card warm up for several minutes so the cooler reaches a steady state.
- Record core, hotspot, and memory temperatures together, then subtract to get the delta.
- Repeat the same test every few months and compare deltas, not just absolute numbers.
Comparing like with like is essential. A delta measured in a lightweight esports title cannot be judged against one from a path-traced game, because the workload distribution across the die is completely different. Ambient room temperature shifts absolute values too, which is another reason the gap between the two sensors is the figure worth tracking.
Fixes, Ranked From Easiest to Most Invasive
When a delta has clearly grown, the fixes form a ladder. The first rung costs nothing: confirm the case airflow is feeding the card cool air and that the fan curve is not tuned so conservatively for silence that the cooler never spins up. An undervolt is next, since trimming voltage reduces heat generation in exactly the dense regions that create the hotspot, and it can shrink both temperatures without opening the card.
Only after those steps does opening the card make sense. Replacing the die paste with a fresh, pump-out-resistant compound addresses the most common cause directly, and inspecting the memory and VRM thermal pads while the cooler is off catches the second most common one. This step carries real risk: warranty seal policies vary by region, and a slipped screwdriver can end a card. Anyone uncomfortable with that trade should stop at the software rungs, which solve a surprising share of cases.
Compounds Suited to a GPU Repaste
GPU dies punish the wrong paste. The die is bare silicon with no heat spreader, temperatures swing hard, and a thin, runny compound that works beautifully on a CPU can pump out of a graphics card within months. The pastes below are the ones builders reach for when delta creep is the problem, chosen for viscosity and stability rather than headline conductivity claims alone.
Noctua NT-H1
A long-running default for graphics card work because it spreads easily, needs no burn-in period, and is formulated for multi-year stability without drying out. It is not electrically conductive, which matters when the application sits millimeters from components around a bare die.
Specs:
– Size: 3.5 g syringe
– Type: Hybrid compound, metal-oxide blend
– Electrically conductive: No
– Burn-in: None required
Noctua NT-H2
The second-generation Noctua compound uses a fine-tuned mixture of metal oxide microparticles for lower thermal resistance and a thinner bond line at typical mounting pressures. The tube ships with cleaning wipes, genuinely useful when stripping old, dried paste off a die and cold plate.
Specs:
– Size: 3.5 g syringe
– Type: Metal-oxide microparticle compound
– Electrically conductive: No
– Included: 3 cleaning wipes
Thermal Grizzly Kryonaut
Built for overclocking, Kryonaut carries one of the highest manufacturer conductivity ratings among conventional pastes and stays stable at high temperatures without curing. The small syringe suits a single careful GPU application, and its thicker consistency resists the migration that causes delta creep.
Specs:
– Size: 1 g syringe
– Thermal conductivity: 12.5 W/mK (manufacturer rating)
– Electrically conductive: No
– Curing: None
ARCTIC MX-6
ARCTIC built MX-6 around carbon filler particles and a noticeably viscous formula aimed at long service life, precisely the profile a graphics card wants. It is neither electrically conductive nor capacitive, removing the short-circuit worry around exposed die-area components, and the 4 g tube covers several applications.
Specs:
– Size: 4 g syringe
– Filler: Carbon-based particles
– Electrically conductive: No
– Capacitive: No
GELID GC-Extreme
A veteran enthusiast compound with a published conductivity figure and a dense consistency that holds its layer under repeated heat cycles. It is non-corrosive, does not cure, and includes an applicator for spreading an even film across the die.
Specs:
– Size: 3.5 g syringe
– Thermal conductivity: 8.5 W/mK (manufacturer rating)
– Density: 3.73 g/cm3
– Operating temperature: -45 to +180 degrees Celsius
How to Choose
For a GPU repaste, prioritize behavior over brochure numbers. Conductivity ratings are measured under conditions that rarely match a real cooler mount, while viscosity and pump-out resistance decide whether the delta stays fixed or creeps back within a season. Quantity matters too: a 1 g syringe is a single application with little margin for a mistake, while 3.5 g and 4 g tubes allow a practice spread and a second mount. Finally, confirm the compound is not electrically conductive before using it near the exposed components around a graphics die.
Who Should Skip This
Skip the whole investigation if the card is new and the delta sits in the normal band, because a 15-degree gap on a fresh card is the design working as intended. Skip it if temperatures only look scary in one lightweight game, since workload differences make single readings meaningless. Skip the repaste if the card is still under a warranty that matters, or if the honest problem is a dusty case and a timid fan curve. And skip worrying entirely if the plan is to replace the card soon anyway; delta creep is a maintenance topic, not an emergency.
Frequently Asked Questions
What is a normal GPU hotspot temperature while gaming?
On most current cards the hotspot lands somewhere between roughly 75 and 95 degrees Celsius in demanding games, depending on the cooler and case. The absolute figure matters less than its distance from the core temperature, and whether that distance has changed over time.
How big a delta between core and hotspot is too big?
A gap of about 10 to 20 degrees Celsius under load is typical for a healthy card. A gap pushing past 25 to 30 degrees under the same workload that used to show less is the pattern that points to paste pump-out, drying, or uneven cooler mounting pressure.
Does a high hotspot mean the graphics card is dying?
Not by itself. A high hotspot with a normal core temperature and a wide delta usually describes a cooling contact problem, not failing silicon. Cards protect themselves by throttling long before the hotspot reaches a damaging level, so the symptom is lost performance and noise.
Can undervolting fix a wide hotspot delta?
It can shrink it. Lower voltage reduces heat generated in the busiest regions of the die, pulling the hotspot down faster than the average. Undervolting will not repair dried or pumped-out paste, but it is the best no-tools step before deciding whether opening the card is worthwhile.
Final Verdict
The hotspot is not a scarier duplicate of the GPU temperature; it is a different instrument, and the gap between the two is the reading that diagnoses cooler health. A stable delta in the teens means the paste and mounting are doing their job, whatever the absolute numbers say. A delta that grows quarter after quarter is the card asking for attention, starting with airflow and voltage long before it asks for a screwdriver.











