How to measure GPU thermals with hot spots and memory junction sensors

GPU temperature is more complicated than the single number shown by many monitoring apps. A graphics card can report a core temperature, a hot spot or junction temperature, and a memory junction temperature, with each sensor describing a different part of the board. Understanding those readings is essential when checking a new card, diagnosing crashes, or deciding whether a second-hand GPU is still in good nick.

The core reading reflects an average across the graphics processor, while the hot spot records the warmest measured point inside the die. Memory junction temperature shows how hot the VRAM is running, which can matter particularly on cards using GDDR6X or high-speed GDDR6. A card may therefore appear comfortable at 70°C while its hot spot or memory is much warmer.

Accurate testing requires a repeatable workload, a known room temperature, and monitoring software that exposes the relevant sensors. Australian conditions make this especially important: a GPU tested in a cool Melbourne room in winter may look very different from the same card running through a Brisbane summer afternoon. Good measurements separate normal environmental variation from genuine cooling problems.

Identify the sensors your GPU exposes

Start with a monitoring tool that can display more than GPU temperature and fan speed. HWiNFO, GPU-Z, AMD Software: Adrenalin Edition and some versions of MSI Afterburner can expose junction values, although the exact labels vary between vendors and models. In GPU-Z, the relevant entries may appear as GPU Temperature, Hot Spot, Memory Temperature or Memory Junction Temperature. HWiNFO often provides additional maximum, minimum and average values.

NVIDIA cards do not all expose the same information. Recent models commonly report a GPU hotspot, while memory junction telemetry depends on the GPU generation and board firmware. AMD Radeon cards frequently provide a clear junction temperature, and many models also report a memory temperature. If a sensor is missing, that does not automatically indicate a fault; the manufacturer may simply not provide it through the driver.

Before testing, record the GPU model, BIOS version, driver version, fan mode and ambient temperature. A basic room thermometer is useful because the temperature inside a closed case can be several degrees higher than the room. Note whether the system is on a desk or under it, and whether the side panel is fitted. Those details make later comparisons much more useful.

Establish a repeatable testing method

Let the computer sit idle for at least ten minutes after booting, with ordinary background applications closed. Record the idle core, hot spot and memory readings, then launch a consistent workload. A demanding game with a built-in benchmark is often preferable to a synthetic stress test because it represents real use, although a tool such as 3DMark or Unigine can provide a repeatable comparison.

Run the test for long enough to reach a stable temperature, usually 20 to 30 minutes. Watch the live values, but pay particular attention to the maximum readings captured during the run. A brief spike can be normal, while a temperature that keeps climbing after several minutes may indicate restricted airflow, an aggressive power limit or a cooler that is not transferring heat properly.

Use the same game settings each time, including resolution, ray tracing, frame-rate limit and upscaling. Uncapped menus can load a GPU harder than gameplay and produce misleading results. For a practical check, a demanding title such as Cyberpunk 2077, Alan Wake 2 or a current racing game can be tested at the resolution the owner actually uses. Hardware Hounds’ video coverage can also provide useful context when comparing visual behaviour, fan noise and workload intensity across cards.

Interpret core, hot spot and memory readings

The GPU core temperature is an average-style measurement and is the figure most commonly shown in a game overlay. Hot spot temperature is different: it identifies the hottest monitored area on the die. It will naturally be higher than the core value, often by 10 to 20°C. A core temperature of 72°C and a hotspot of 88°C may be perfectly healthy, especially in a compact case or during a heavy render.

The gap between core and hot spot is called the delta. A consistently large delta can suggest uneven cooler contact, dried thermal paste, mounting pressure problems or a heatsink that is not sitting squarely. There is no universal failure number, because cooler design, power draw and firmware behaviour all matter. A rising delta above roughly 25 to 30°C is worth investigating, particularly if the card is relatively new or has recently been moved.

Memory junction requires a separate interpretation. VRAM can run hotter than the GPU core, and memory-heavy workloads may increase it without producing a dramatic change in core temperature. High-resolution textures, ray tracing, cryptocurrency workloads and professional rendering can all place extra demand on memory. Compare the result with the manufacturer’s specified operating limits where available, rather than applying a single temperature rule to every graphics card.

Test airflow, noise and Australian conditions

Thermal testing should include fan behaviour and noise. A card that reaches a safe temperature by running its fans at 2,000 rpm may be less pleasant than one that sits a few degrees warmer at a much quieter speed. Record fan percentage or rpm alongside the temperatures, and note whether the fans stop at idle. A sudden fan ramp, rattling bearing or cyclical speed change can point to a control issue even when temperatures look acceptable.

Case airflow has a strong effect in Australia. During a hot Sydney or Perth day, an indoor ambient temperature around 30°C can push load temperatures substantially higher than a winter test. Brisbane users may also be dealing with humid conditions and rooms that retain heat overnight, while dusty environments in regional areas can clog front filters and heatsinks quickly. Testing with the same side-panel position and fan profile gives a fairer picture than comparing isolated screenshots.

A practical airflow check is to repeat the workload with the case side panel closed, then briefly with it open. If the core and hotspot both fall by a large amount with the panel removed, the case is probably starved of cool intake air or is exhausting poorly. Clean filters, confirm that front fans are intake and rear or top fans are exhaust, and check that cables are not blocking the main airflow path. Australian power costs also make sensible fan curves worthwhile: the goal is stable performance without needless high-speed cooling.

Diagnose abnormal readings and make a useful record

When a GPU runs hotter than expected, check the simple causes first. Confirm that the card is not being overclocked, that the fans respond to load, and that the heatsink is free of dust. Inspect the case for poor intake clearance, especially if it sits in a desk nook. Also check whether a neighbouring expansion card is restricting the GPU’s intake or whether a vertical mount places the fans too close to the side panel.

A high memory junction temperature with a normal core can indicate limited VRAM cooling or worn thermal pads. A very high hotspot delta with a reasonable average core temperature points more towards die-to-cooler contact. Replacing paste or pads can help, but it may void a warranty and requires the correct thickness and compression. For a new card, an RMA or retailer assessment is usually safer than opening it immediately. Australian buyers should keep invoices and test results because warranty handling can differ between a local retailer, an importer and a marketplace seller.

Keep a short log containing ambient temperature, workload, duration, resolution, power draw, core temperature, hotspot temperature, memory junction temperature, fan speed and noise observations. Screenshots from HWiNFO or GPU-Z are useful, but a written record makes trends easier to follow. When checking community reports or broader hardware resources, compare exact GPU models and cooler designs rather than assuming every version of a chipset behaves identically.

A sensible result is one that remains stable without sustained thermal throttling, excessive fan noise or an unusually large hotspot delta. Use the recorded evidence to decide whether the card needs better case airflow, a revised fan curve, lower voltage, a power limit adjustment or professional servicing. For readers comparing new and used hardware, the combination of core, hotspot and memory readings gives a far more reliable picture than a single green temperature number on a gaming overlay. Further testing guides and component coverage are available through the Hardware Hounds website.