NVIDIA GPU Overclocking

Guide for tuning NVIDIA GPUs with MSI Afterburner, focusing on core clock, undervolting, and memory overclocking.

Overclocking raises clock speed and usually increases temperature. Undervolting caps voltage so the GPU can run with lower wattage and temperature while keeping useful performance.

Recommended order

1. Install monitoring and tuning tools.
2. Configure MSI Afterburner, fan curve, and power limits.
3. Set up benchmark and stress tools.
4. Find a stable core clock and voltage.
5. Undervolt if thermals or power limits are the bottleneck.
6. Tune memory after core settings are stable.
7. Run final validation.

Before you start

Install HWiNFO and run it in sensors only mode during stress tests.

This guide skips VBIOS flashing. If you still want to try it, use the correct VBIOS and follow a dedicated flashing guide first.

During tuning, watch the performance limit in HWiNFO. It should not stay at Yes all the time.

Power limit

If the card constantly hits the power limit, it downclocks. Raising clock alone will not help until voltage and power headroom are under control.

Tools to install

Tool	Role in this guide
MSI Afterburner	Fan curve, voltage curve, power limits, and profiles
Superposition	Core and VRAM stress during tuning
3DMark	Additional benchmark runs while finding core clock
OCCT	Error checking after OC/UV is complete
memtest vulkan	Final VRAM stability pass
FurMark	Optional heavy power test at the end

MSI Afterburner setup

Fan curve

Set a custom fan curve and keep fan speed high enough that thermals do not skew your results. You can import the preconfigured cfg file or configure it manually.

Load a profile at startup

If you do not want MSI Afterburner running in the background all the time, set a static curve and load a profile at startup:

schtasks /create /sc ONSTART /tn "MSIAfterburnerProfile" /tr "powershell.exe -NoProfile -Command \"Set-Location 'C:\Program Files (x86)\MSI Afterburner'; .\MSIAfterburner.exe /profile1\"" /rl HIGHEST /delay 0000:20

Parameter	Description
/sc	Schedule type. ONSTART runs the task every time the system starts.
/tn	Task name. Each scheduled task must have a unique name.
/tr	Program or command to run. Use the full path to the executable.
/rl	Run level. HIGHEST runs with elevated privileges.
/delay	Wait time after trigger in mmmm:ss format.

See schtasks create for the full parameter list.

Power and temperature limits

Set Power Limit and Temp. Limit to their maximum values, change priority to power limit, and disable automatic startup for now so a bad profile does not loop on launch.

Benchmark and stress settings

3DMark

Use Time Spy and Steel Nomad. Avoid Wildlife, Solar Bay, and light versions.

Superposition

Use the highest preset that does not exceed your VRAM limit to test core clock. Use 8K or 4K Optimized to stress VRAM.

OCCT

Open the 3D Adaptive tab and use these settings:

When to use OCCT

Use OCCT after core and memory tuning are done. Avoid it while searching for voltage, because it often pushes the card into throttling.

Core overclocking

Find a voltage and clock combination that does not constantly hit the performance limit, downclock, or crash.

• If a stress test crashes at +180 @ 1025 mV, +180 is unlikely to be stable at a very different voltage.
• Use multiple stress tests and run each for about 30 min, not just a few seconds.
• Core clock and effective clock should stay close together.

Architecture	Step size	Examples
Kepler, Maxwell, Pascal	12.5 MHz	GTX 600/700/900, GTX 10 Series, Titan cards
Turing, Ampere, Ada Lovelace	15 MHz	RTX 20/30/40/50 Series

Core clock and voltage

1. Start at 900–950 mV with +50 MHz to +75 MHz core offset and run Superposition or 3DMark once.
2. Watch temperatures in HWiNFO during the run. The GPU should warm up above 75°C for a meaningful test. If you are already near ~85°C, skip overclocking and undervolt instead.
3. If the run finishes cleanly, increase core clock by 15 MHz or 12.5 MHz and test again. You can also raise voltage if you are not hitting the power limit yet.
4. Repeat until one of these happens:
   • Crashes: lower core clock.
   • Power limit: lower voltage.
5. You should end with a voltage and core clock that pass stress tests without constant crashes or power-limit downclocking.

Example result:

Undervolting

Undervolting limits GPU voltage so wattage and temperature drop while keeping usable performance. Use it when thermals or power limits stop you from holding a useful core clock.

Limit the voltage curve

1. Open the curve editor (Ctrl + F).
2. Click the next point to the right of the millivolt value you want, for example 906 if you want 900 mV.
3. Hold Ctrl and select the curve from the right up to that point.
4. Drag the selected point down so the last point ends below the first point, for example 900 mV.

5. Save the settings to a profile.

Use the same curve-editor steps during core tuning whenever you need to cap voltage at a specific millivolt value.

Memory overclock

Use a stable core clock first. Save benchmark results before tuning memory, because you will often stop when scores get worse even if the card does not crash.

Watch for artifacts such as flickering textures, colored pixels, or screen tearing. Most cards can take a large memory offset, so you can start around 250–500 MHz.

Increase the memory clock

1. Enter your starting memory clock.
2. Stress test with Superposition.
3. If stable, increase memory clock by 100 MHz.
4. Repeat until scores drop, artifacts appear, or the run crashes.
5. Go back to the last stable value and add 50 MHz.
   • Stress test again. If stable, increase by 5–10 MHz. If not, drop 50 MHz and tune up in 5–10 MHz steps.
6. Save the stable memory clock to a profile.
7. Validate VRAM with memtest vulkan for about 30–60 min.

Final validation

Run these checks after core, undervolt, and memory settings are saved:

• OCCT look for errors and confirm OC/UV stability.
• FurMark optional heavy power test; run it briefly after the other steps are complete.