HWiNFO64 GPU Monitoring Guide: The Sensors That Actually Matter

HWiNFO64 lists over 50 GPU-related sensors for a typical discrete graphics card. Most are noise. A focused set of about a dozen sensors gives you a complete picture of what your GPU is doing during an overclock, an undervolt test, or a sustained gaming session.

When you open HWiNFO64 in Sensors-only mode and scroll to the GPU section, the list is overwhelming. There are temperatures, clocks, voltages, power readings, percentages, and counters, often with names that overlap or differ only in subtle wording. Knowing which sensors to watch and what they mean converts HWiNFO64 from a wall of numbers into a precise diagnostic instrument.

Setting up HWiNFO64 for GPU monitoring

Download HWiNFO64 from hwinfo.com. Install or run the portable version. When it launches, select Sensors Only to skip the hardware summary screen. The sensor list will open automatically. Scroll down to the section labeled with your GPU model. Expand it if it is collapsed.

To reduce clutter, right-click any sensor value and choose Show Graph to open a live chart, or check the box next to sensors you want to watch. You can also right-click a sensor and select Add to Tray to pin it as a system tray icon for monitoring during a fullscreen game. For logging, press the logging button (floppy disk icon) to start a CSV log—useful for reviewing peak and minimum values after a benchmark run.

GPU Core Clock

This is the actual operating frequency of the GPU shader core at this moment. It fluctuates constantly. During gaming it usually sits within a 30 to 50 MHz band that represents the GPU's sustained boost clock under current power and thermal conditions. If this number is significantly lower than your GPU's rated boost clock during a demanding workload, the card is throttling.

After applying an overclock in MSI Afterburner, this sensor confirms whether the change is active. If you set a curve ceiling of 2800 MHz and the Core Clock sensor reads 2750 MHz under load, that is normal boost variation. If it reads 2600 MHz, either the power limit is constraining it or thermal throttling is occurring—check the next two sensors to determine which.

GPU Core Voltage

This shows the actual voltage the GPU is operating at in real time. Compare this to your voltage/frequency curve settings in Afterburner to verify the curve is being applied. A GPU that is flatlining at maximum voltage (1000 mV on many Ada Lovelace cards) while also not reaching its clock target is power-limited—it is requesting maximum resources and still cannot maintain the target frequency. A GPU that is running at lower-than-expected voltage is either correctly following a reduced-voltage curve or is being throttled by the driver.

GPU Power (W)

This reports total board power consumption in watts. For desktop GPUs this is the aggregate of PCIe slot power and all PCIe power connector draws. The reading is critical for two reasons. First, it tells you how close to the power limit you are operating. If your RTX 4080 has a 320W power limit and is drawing 318W sustained, it is power-limited. Raising the power limit slider in Afterburner may unlock more performance. Second, after an undervolt, a lower power reading at the same clock speed confirms the undervolt is effective and the GPU is doing the same work with less energy.

GPU Temperature and GPU Hot Spot Temperature

Modern NVIDIA cards report two temperature sensors. GPU Temperature is the average die temperature across the full surface of the GPU package, measured by a central diode. GPU Hot Spot Temperature is the highest temperature detected by any of several distributed thermal sensors across the die. The hot spot reading is always higher than the average—on a well-cooled RTX 4070, you might see 70 C average with an 84 C hot spot under sustained load.

Hot spot temperature is the number that matters for throttling decisions. NVIDIA cards typically begin thermal throttling when the hot spot approaches 83 to 87 C depending on the model. If your core clock is dropping under load and your hot spot is in this range, thermal throttling is the cause. The fix is better airflow, repaste, or a reduced voltage/frequency curve that generates less heat.

GPU Memory Junction Temperature

For GDDR6X-equipped cards (RTX 3080 and later high-end NVIDIA), this sensor reports the temperature of the VRAM junction—the hottest point on the memory dies. GDDR6X regularly exceeds 100 C during demanding workloads and is rated to operate up to 110 C. Sustained operation at 105 to 108 C is not a failure condition but it does indicate that memory throttling may occur at sustained gaming loads. GDDR6X throttles its data rate when the junction temperature exceeds around 105 C on most implementations.

If your VRAM junction temperature is consistently above 100 C, improving case airflow or underclocking the memory (which reduces heat from the memory itself) can bring it into a range that avoids memory throttling.

GPU Throttling Sensors

HWiNFO64 includes binary throttling sensors that show Yes or No states for different throttle reasons. The most useful are: Thermal Throttle (GPU temperature has exceeded the thermal limit), Power Throttle (power draw has hit the power limit), Voltage Throttle (the GPU is clipping at a voltage ceiling), and Current Throttle (VRM current limits are active). During a gaming session where performance feels inconsistent, log all throttle sensors and review which ones flip to Yes during the performance dips. This directly identifies the constraint without guesswork.

GPU Fan Speed

Reports fan RPM. Cross-reference this with GPU temperature to verify the fan curve is working. A GPU at 85 C with fans at 40% speed indicates the fan curve is not aggressive enough. A GPU at 65 C with fans at 100% speed indicates a cooling problem unrelated to fan effort—typically thermal paste failure or blocked heatsink fins. HWiNFO64 reports fan speed as both RPM and percentage of maximum for most GPU models.

GPU Memory Clock

This shows the actual operating frequency of the VRAM. For GDDR6X cards this is typically reported as the transfer rate in Gbps or as the base clock in MHz. During idle or light use the memory downclock significantly to save power. Under gaming load it should be at or near its rated frequency. After applying a memory overclock in Afterburner, this sensor confirms the new frequency is active. A memory overclock that drops the memory clock below rated frequency under load indicates the overclock is unstable and the driver is resetting the memory to a safe frequency.

Combining sensors into a tuning workflow

Effective GPU tuning requires watching these sensors together, not in isolation. The typical sequence for validating a voltage/frequency curve change is: apply the change in Afterburner, run a sustained benchmark for 20 minutes, log HWiNFO64 data throughout, then review the log. Check that GPU Core Clock held near your target, GPU Power stayed below the limit (or hit it in a predictable way), GPU Hot Spot Temperature stayed below throttle threshold, and no throttle sensors activated. If the benchmark score is also consistent across runs, the tuning is stable.