MSI Afterburner Overclocking Guide: Safe Settings for NVIDIA GPUs

Short answer: install MSI Afterburner, raise the power and temperature limits to maximum, set a firmer fan curve, then add core clock in +15 MHz steps and memory in +50 to +100 MHz steps, stress-testing 10 minutes after each change. Most NVIDIA GPUs land at +100 to +150 MHz core and +500 to +1000 MHz memory.

What MSI Afterburner Actually Changes

Afterburner does not unlock your GPU. GPU Boost already runs the chip as fast as its power, temperature and voltage budget allow. Afterburner gives you four dials that shift those budgets:

• Core Clock (MHz) — an offset applied across the whole boost curve, not a fixed frequency. +100 means every boost step tries to run 100 MHz higher.
• Memory Clock (MHz) — an offset to the VRAM clock, usually where the free performance is on memory-bound cards.
• Power Limit (%) — how much wattage the card may pull before throttling itself.
• Temp Limit (°C) — the thermal ceiling at which it starts dropping clocks.

Because the core slider is an offset, a “+150 MHz overclock” on a card that already thermal-throttles does nothing at all. Fixing cooling often beats fixing clocks, which is why the fan curve comes before the clock steps here.

Step 1: Install Afterburner and Set Up Monitoring

Download Afterburner only from MSI’s own site — the search results for it are a minefield of bundled installers. The 4.6.6-series builds support current RTX 50-series cards; older 4.6.4 builds install but may not expose every slider on new silicon.

Keep RivaTuner Statistics Server (RTSS) checked during install, because it draws the on-screen overlay and without live numbers you are overclocking blind. Under Monitoring, enable “Show in On-Screen Display” for GPU temperature, hot spot / memory junction temperature, core clock, memory clock, power percentage and framerate. Leave “Start with Windows” off until you have a profile you trust — auto-applying an untested overclock at boot is how people end up in a reboot loop.

Step 2: Raise Power and Temperature Limits First

Before touching a single MHz, drag Power Limit to maximum and Temp Limit to roughly 83-87°C, then apply. Both are effectively free: you are not changing voltage, only telling the card it may use headroom its own firmware already considers safe.

Headroom varies enormously. Some RTX 40 and 50 cards ship with a locked +0% power slider; others allow +110% to +133%. A slider that will not move is a BIOS limit, not a bug — focus on memory clocks and cooling instead. Run a benchmark now and note the score: without that baseline you cannot tell whether a later setting helped. If the machine is new, check our notes on common first-build mistakes first.

Step 3: Build a Fan Curve and Keep Temperatures in Check

Open the gear icon, go to the Fan tab, tick “Enable user defined software automatic fan control”, and shape a curve. A sane starting shape: 30°C at 0-30%, 50°C at 40%, 65°C at 60%, 75°C at 80%, 83°C at 100%. Set the update period to 2000-5000 ms so fans do not surge during loading screens.

If the card sits above 80°C at stock with fans past 70%, the bottleneck is case airflow, not the GPU. Two front intakes and a rear exhaust typically drop GPU temperature several degrees and make every later step land better. Our guide to safe GPU temperature ranges explains what hot spot and memory junction readings mean.

Step 4: Overclock the Core Clock in Small Steps

Move the Core Clock slider to +50 MHz, apply, and run a 10-minute stress test. If it passes, add +15 to +25 MHz and repeat. Do not jump straight to +200; when it crashes you will not know which value was the last good one. Keep a text file of every offset and its result — the process is a binary search and falls apart without notes.

When you hit an artifact or a crash, drop back 30 MHz below the last value that passed, not to it. Ten minutes of stability is not forty hours of stability, and that margin absorbs a hot afternoon.

A note on the curve editor

Ctrl+F opens the voltage/frequency curve editor, which is how undervolting is done: pick a point around 900-950 mV, drag it up to your target clock, flatten everything to its right, apply. Undervolting often delivers near-overclock performance at 30-60 W less power, and on gaming laptops and small-form-factor builds it beats a straight overclock.

Step 5: Overclock the Memory Clock

Memory can move in bigger jumps: +200 MHz at a time until you are close, then +50 MHz. Bandwidth-limited cards at 1440p and 4K often gain more here than from core.

The trap: GDDR6X and GDDR7 use error correction that silently retries failed transfers, so instead of crashing the card quietly loses performance. The memory test is therefore not “did it crash” but “did the score go up”. If a +200 MHz step drops your score, you are already past the useful limit.

Safe Starting Points by GPU Generation

These are typical community-reported ranges, not guarantees. Silicon varies card to card, and two identical models can differ by 100 MHz. Treat them as where to begin, not where to end.

How to Stress-Test Each Step

No single test catches everything, so a pass in one tool means little on its own.

• Unigine Heaven / Superposition — fast and visual, the best tools for spotting artifacts: flickering triangles, sparkles, stray coloured dots. Superposition’s 1080p Extreme preset makes a good loop.
• OCCT — the important one for memory. Its VRAM and 3D adaptive tests check for computation errors and report a silent error count.
• 3DMark Steel Nomad or Speed Way stress test — reports a frame-to-frame stability percentage, roughly 97% being the usual pass mark.
• Actual games, for an hour — the final exam. Shader-heavy and ray-traced titles find instability synthetic loops miss.

FurMark is a power virus that forces protective throttling, so it tests cooling rather than clocks — use it to validate the fan curve only. For more, see our rundown of GPU benchmarking tools.

Troubleshooting: Telling a Crash from Instability

The failure mode tells you which slider to blame.

• Black screen then recovery (“display driver stopped responding”) — core clock too high. Drop the core offset 30 MHz.
• Artifacts with no crash — memory, almost always. Drop the memory offset 100-200 MHz.
• Score falling as you add memory clock — error correction working overtime; you are past the limit despite no crash.
• Hard lock or instant reboot — power delivery. Confirm your power supply has enough headroom before blaming the GPU.
• Stable in benchmarks, crashes in games — a marginal overclock. Take another 15-30 MHz off the core.
• Clocks dropping despite a high offset — not instability, but the temperature or power limit. More offset will not help.

How to Revert

Nothing here is permanent. The circular-arrow reset button returns every slider to stock instantly, and closing Afterburner without saving does the same. If it applies at startup and the machine will not boot cleanly, use Safe Mode to uninstall it or delete its profile files. Save settings to numbered profile slots so an everyday profile and a benchmark profile can coexist.

Is Overclocking Safe? The Honest Version

Overclocking through Afterburner is low-risk because the sliders work inside limits the card’s firmware enforces. You cannot dial in a dangerous voltage — the core voltage slider is capped, and locked entirely on many models — and modern GPUs shut down before thermal damage. The real risks are mundane: crashes that lose unsaved work, and instability that surfaces weeks later in one game. BIOS flashing is a different risk category entirely.

Be realistic about the payoff. A good GPU overclock is typically 3-8% more performance — 60 fps becomes about 63-65 fps. It will not rescue a card a tier too slow for your monitor, so if you are chasing high refresh rates, checking the right GPU tier for 1440p is a better use of an evening.

Frequently Asked Questions

Will MSI Afterburner void my GPU warranty?

Software overclocking with Afterburner is generally not grounds for a warranty denial, and it is not practically detectable afterwards because the settings live in Windows rather than on the card. Policies vary by manufacturer. Flashing a modified video BIOS is entirely different: it writes permanently to the card and voids coverage with most vendors.

What is a safe GPU temperature while overclocking?

Under sustained load, keep the core below 80°C, with up to roughly 83-85°C acceptable on cards that run hot by design. Memory junction temperature matters separately on GDDR6X models: readings approaching 105°C mean the VRAM is throttling, so improve airflow before adding memory offset. Hot spot normally sits 10-15°C above core.

Should I overclock or undervolt my NVIDIA GPU?

If your card is power-limited or thermally limited, which covers most RTX 40 and 50 cards, laptops and small-form-factor builds, undervolting is the better choice. It frequently matches near-overclocked performance while pulling 30-60 W less and running cooler and quieter. Pure overclocking suits older, well-cooled desktop cards with genuine power-limit headroom left.

How long should I stress-test a GPU overclock?

Run about 10 minutes per incremental step while tuning, which catches obvious failures quickly. Once the settings look final, validate with roughly 30 minutes of mixed synthetic load plus at least one full hour in a demanding game. Games stress shader paths and memory access patterns that synthetic loops never touch, and marginal overclocks usually fail there first.

Why does my GPU not reach the clock speed I set in Afterburner?

Because the core slider is an offset to the boost curve, not a fixed target. GPU Boost still cuts clocks whenever it hits the power limit, temperature limit or a voltage ceiling, so a hot or power-starved card can run at or below stock despite a large offset. Check the limit indicators in your overlay and fix the active limiter first.