Published August 5, 2026 - Updated August 5, 2026 - By Xilly
NVIDIA GPU Overclock and Undervolt Guide
Find your NVIDIA GPU's stable core and memory ceiling first, then reduce voltage while preserving as much of that performance as possible.
GPU Optimization Guides PC Optimization Guides
NVIDIA GPU Tuning ยท Overclock + Undervolt
Overclocking and undervolting are not opposites. The useful order is simple: first find the highest clock speed your graphics card can hold reliably, then lower the voltage while trying to preserve that performance. Done carefully, this can improve FPS, reduce power draw, lower temperatures, and cut fan noise at the same time.
This custom 3DMark package contains the Steel Nomad benchmark used throughout the guide without the rest of the roughly 14 GB 3DMark suite. The download is about 2 GB, so you get the benchmark you need without installing every other test.
Download, extract the folder, run the included 3DMark setup, and use the same Steel Nomad test for every core-clock comparison.
GPU tuning is never guaranteed. Every card is different, including two cards with the same model name. Save important work, monitor temperatures, change one variable at a time, and keep a stock profile available. Do not blindly maximize the core-voltage slider. Unlocking voltage monitoring can help you read and edit the curve, but unnecessary voltage adds heat and risk.
Why Overclock First and Undervolt Second?
An overclock finds the performance ceiling of your individual card. An undervolt then looks for the lowest voltage that can sustain a useful point near that ceiling. If you undervolt before learning what the card can actually do, you have no reliable performance target and no clean baseline for comparison.
Run Steel Nomad at stock settings and save the score, clock speed, temperature, and power draw.
Tune memory and core in controlled increments, testing after every meaningful change.
Flatten the voltage-frequency curve at a tested point, then compare performance and stability again.
Set Up MSI Afterburner
Install MSI Afterburner from its official source. RivaTuner Statistics Server is optional and is not required for this process. Open Afterburner, click the settings gear, enable voltage control and voltage monitoring if your GPU supports them, then restart the app if prompted.
Set the power limit to the highest value the card's firmware allows so power limiting does not distort the baseline. You can link or raise the temperature limit if appropriate, but keep an eye on actual temperatures throughout testing. Leave the core and memory offsets at zero for the first run. The goal is to remove an artificial power ceiling, not to add voltage immediately.
Run and Record a Stock Steel Nomad Baseline
Close unnecessary background apps and run Steel Nomad once at stock core and memory settings. Record the score. Also note the sustained GPU clock, peak temperature, and approximate board power if you monitor them. This is the control result that every later run must beat or meaningfully match.
Use the same benchmark, resolution, driver, background apps, and room conditions. A score change means much less if Windows Update, a browser, or a different test was running in the background.
Find a Stable Memory Overclock
Memory tuning is easiest to isolate before the core clock changes. Open MemTest Vulkan only when you are ready to test because it begins stressing VRAM immediately. Start with a conservative memory offset, apply it, and watch for reported errors, visual corruption, driver resets, or crashes.
A benchmark score can also expose memory instability. Error correction may prevent an obvious crash while silently reducing performance. If the score stops increasing or falls after a higher memory offset, the lower setting may be faster in practice.
Find the Maximum Stable Core Clock
With the memory setting applied, begin around a +75 MHz core offset. Run Steel Nomad and compare the score with your baseline. If it completes cleanly and the score improves, increase the core by 15-30 MHz and test again.
- 1Start at +75 MHz
This is a reasonable test point for many modern cards, but lower it if your card crashes or behaves abnormally.
- 2Increase by 15-30 MHz
Run the same Steel Nomad benchmark after each change and write down every score.
- 3Stop when performance stalls
A crash is an obvious limit. A flat or lower score can also mean the card has stopped gaining useful performance.
- 4Back off for daily use
Reduce the last unstable core setting by 15-30 MHz, then leave additional headroom if real games expose instability.
Save the stable overclock as Profile 1. This gives you a known performance profile and an easy recovery point before you begin editing the voltage-frequency curve.
Undervolt the Stable Overclock
Open the voltage-frequency curve editor in MSI Afterburner with Ctrl + F. Pick a voltage point to test. A starting point around 1000 mV is often useful for a high-performance tune, but it is not a universal target.
Select the curve points to the right of the chosen voltage and flatten them so the GPU does not request a higher frequency above that voltage. Apply the curve, run Steel Nomad, and compare the result with the stable overclock profile.
If 1000 mV is stable, you can try a lower point such as 975 or 950 mV. Expect the available frequency to fall as voltage drops. The goal is not the lowest possible voltage. The goal is the efficiency point that fits your PC and your priorities.
Validate in Real Games
Steel Nomad and VRAM testing are fast filters, not final proof. Save the undervolt as a second profile and test several games for a few hours. Include both demanding GPU-heavy titles and lighter games where the card boosts to high clocks. Different engines expose different failures.
Watch for crashes, black screens, flashing or corrupted textures, visual artifacts, driver-reset notifications, sudden score regression, or clock-speed drops. Lower the core by 15-30 MHz or memory by 100 MHz, then repeat the relevant stress test and gaming session.
What Kind of Improvement Should You Expect?
The result depends on the GPU, cooling, silicon quality, game, and whether the stock card was already near its limit. In the example from this guide, Steel Nomad moved from 9,183 at stock to roughly 9,827 after tuning. In a path-traced Cyberpunk test, power draw fell from roughly 380-400 watts to around 320-330 watts while performance improved slightly.
Those numbers are examples, not promises. GPU-heavy games may show a modest FPS gain, while titles with more headroom can gain more. If your game is limited by the CPU, memory, Windows setup, or frame cap, a GPU overclock cannot remove that bottleneck.
Quick NVIDIA GPU Tuning Checklist
A GPU Tune Is Only One Layer
A GPU overclock cannot fix bad Windows settings, background bloat, incorrect BIOS configuration, unstable RAM, or poor game settings. If the card is tuned but the PC still stutters or feels delayed, the bottleneck is probably somewhere else in the system.
Professional PC Optimization
Want the Entire PC Tuned?
Xilly tunes the full stack for your hardware: BIOS, RAM, Windows, drivers, GPU behavior, and game settings. More than 10,000 PCs optimized.
Frequently Asked Questions
Should I overclock or undervolt my NVIDIA GPU first?
Overclock first to measure the card's stable performance ceiling. Then undervolt while trying to preserve a useful clock near that ceiling. This gives you a clear target and makes the efficiency trade-off measurable.
Can undervolting increase FPS?
Sometimes. Lower voltage can reduce temperature and power limiting, allowing the GPU to hold its boost clock more consistently. The outcome depends on the card and workload, so compare scores instead of assuming a gain.
Is Steel Nomad enough to prove my overclock is stable?
No. It is a useful repeatable benchmark, but final validation requires multiple real games and longer sessions. Different workloads can expose instability that one synthetic test misses.
How do I know when GPU memory is unstable?
Look for VRAM-test errors, artifacts, flashing textures, crashes, driver resets, or benchmark scores that stop improving. Memory error correction can reduce performance without producing an obvious crash.
What voltage should I use for an NVIDIA GPU undervolt?
There is no universal voltage. Around 1000 mV can be a practical first test for a high-performance tune, but the correct point depends on the exact GPU, clock target, cooling, and stability results.