How to Overclock a GPU: The Complete 2026 Guide

Prep Your System and Record a Baseline Benchmark

Before changing a single setting, install the current stable driver from NVIDIA or AMD's website (not a beta), and update your overclocking software to its latest version — old builds sometimes mismatch newer GPU firmware and report incorrect voltage or power figures. Close background apps that use the GPU, like browsers with hardware acceleration and streaming/recording software.

Next, run a benchmark before you touch anything so you have a number to compare against later. Good free options are 3DMark's Time Spy (NVIDIA and AMD), Unigine Superposition, or Unigine Heaven. Run the benchmark three times at stock settings and note the average score, minimum frame rate, and peak temperature — this is your baseline. Skipping this step is the single most common overclocking mistake: without a baseline, you have no way to tell whether your later changes actually helped, hurt, or did nothing.

While you're at it, open your overclocking software's monitoring graphs (Afterburner's OSD or Adrenalin's Performance Metrics overlay) and note your stock core clock, memory clock, GPU temperature, hotspot/junction temperature, and power draw under load. You'll reference these numbers constantly over the next few steps.

What you'll need before starting:

• The latest stable (WHQL) driver for your GPU, plus your overclocking software of choice installed and updated
• A benchmark or two — 3DMark, Unigine Superposition/Heaven, or a demanding game you own — to establish and later re-check performance
• A monitoring overlay running (Afterburner's OSD, Adrenalin's metrics overlay, or a standalone tool like HWiNFO64) so you can watch clocks, temperatures, and power draw live
• Roughly an hour of uninterrupted time — rushed overclocking sessions are how people skip the stability testing step and end up with a card that crashes days later under a slightly different workload
• A sense of your PSU's rated wattage and your system's approximate total draw, so you know how much headroom you have before raising the power limit in Step 4

Increase the Core Clock Offset

The core clock offset is the setting with the biggest effect on in-game frame rate. In MSI Afterburner or EVGA Precision X1, this is the "Core Clock" slider (shown in MHz above 0); in AMD Adrenalin, it's under Performance → Tuning → Manual → GPU tab.

Raise the core clock offset in small increments — +10 to 15 MHz for NVIDIA cards, +25 to 50 MHz for AMD cards, which tend to have coarser clock-state granularity. After each increase, click "Apply" and run a short stress test (5-10 minutes of Superposition or the built-in AMD Adrenalin stress test) to check for artifacts: flickering textures, colored triangles or squares on screen, driver-crash popups, or a black screen followed by a display recovery message. Any of those means you've gone one step too far — dial the offset back down by the last increment and that's your stable core clock ceiling for now (you'll revisit it after setting power limit and voltage in Step 4).

Most modern NVIDIA cards land somewhere around +100 to +200 MHz core offset before hitting instability; AMD RDNA cards are often more clock-limited and land around +50 to +150 MHz. Don't chase a specific number — chase the point right before instability, since silicon quality ("the silicon lottery") varies card to card even within the same model, meaning two identical GPUs straight off the same production line can top out at noticeably different offsets.

A faster way to find your rough ceiling, if you don't mind a slightly less methodical first pass: jump in larger increments (+25-30 MHz for NVIDIA, +50-75 MHz for AMD) until you hit instability, then back off by roughly double your normal step size and switch to the fine 10-15 MHz / 25-50 MHz increments from there. This gets you to the interesting range faster without sacrificing the precision that matters near your actual limit. Either approach works — the fine, cautious version described above is simply lower-risk for a first-time overclocker who isn't yet used to reading artifact warning signs.

Increase the Memory Clock Offset

With your core clock offset locked in from Step 2, move to the memory clock offset — the same software panel, a separate slider usually labeled "Memory Clock" or "VRAM." Memory overclocking tends to have a wider stable range than core clock and often delivers a real gain in memory-bandwidth-limited titles and resolutions (1440p and 4K especially).

Increase memory offset in steps of +25 to 50 MHz, re-testing after each bump the same way you did for core clock. Memory instability shows up a bit differently — instead of artifacts, you're more likely to see stuttering, a driver-timeout error ("Display driver stopped responding and has recovered"), or a full crash to desktop. GDDR6/GDDR6X memory on NVIDIA cards can often take large offsets (+500 to +1,000+ MHz is not unusual on high-end cards), while GDDR6 on AMD cards is typically more conservative — often +100 to +300 MHz. Back off by one increment the moment you see any instability, the same rule as core clock.

Note that Afterburner and Precision X1 report the memory offset differently than the card's actual effective data rate — a +1,000 MHz offset on GDDR6X, for example, doesn't mean the memory jumps by a literal 1,000 MHz in real clock terms, since GDDR reports and applies offsets at twice the effective transfer rate. Don't worry about the underlying math; just keep using the software's own offset number as your reference point and testing methodically the same way regardless of what the raw figure looks like. AMD Adrenalin displays memory clock more directly in its Tuning tab, which is one reason AMD overclocking tends to feel more approachable to first-timers.

Adjust the Power Limit and Voltage

Your core and memory offsets from Steps 2-3 are capped by how much power and voltage the card is allowed to draw. Raising the power limit slider — usually expressed as a percentage of the card's rated TDP, up to a hard cap around 110-120% on most consumer cards — gives your GPU headroom to sustain its boost clock under sustained load instead of throttling down. Move this slider to its maximum first if your case airflow and PSU headroom support it (check your power supply's rated wattage against your system's total draw before doing this), then re-run Step 2's core clock test; you'll often be able to push a few more MHz once power limit is no longer the bottleneck.

Voltage is the more advanced lever. Afterburner and Precision X1 expose a voltage/frequency curve editor: click-and-drag points on the curve to either raise voltage at a given clock speed (pushing for higher sustained boost) or, more commonly recommended for most users, flatten the curve to undervolt — running the same or slightly higher clock speed at a lower voltage, which cuts heat and power draw with little to no performance loss. AMD Adrenalin offers a similar manual voltage offset under the same Tuning tab. If you're new to overclocking, we recommend leaving voltage at its default (or applying a mild undervolt) rather than manually raising it — voltage increases are where most of the long-term wear risk lives, and the performance gain from added voltage is usually a couple of percent at best.

If you do want to try undervolting specifically, the process is the reverse logic of the core clock test in Step 2: instead of raising the offset until you see instability, you lower the voltage at your target clock point on the curve until the card either drops out of that clock state or shows instability, then move it back up one notch. A well-tuned undervolt commonly cuts power draw by 15-25% and drops load temperatures by 5-10°C, with performance staying within a percent or two of stock — which is why many enthusiasts consider undervolting the single best "overclock" for anyone who games with the case side panel off or in a small form factor build where noise and heat matter more than the last few frames per second.

Run a Full Stability Test

A 5-10 minute check between slider adjustments catches obvious instability, but it won't catch everything — some overclocks pass a short test and then crash 40 minutes into a gaming session once the card has fully warmed up. Once you've settled on core clock, memory clock, and power limit values you like, run a dedicated stress test for at least 20-30 minutes continuously. Good options:

• 3DMark Time Spy Stress Test — runs the same test loop repeatedly and reports a consistency percentage; anything below 97% suggests thermal throttling or instability
• FurMark 2 — an intentionally extreme power-virus-style load; useful for finding your absolute worst-case temperature and power draw, though it's harsher than any real game
• OCCT (GPU:3D or VRAM test) — includes automatic error detection that flags the exact moment memory or core instability occurs
• Unigine Superposition/Heaven looped for 30+ minutes — a more "real-world" load than FurMark

If the stress test completes without a crash, artifact, or driver-timeout error, your overclock is provisionally stable. We still recommend a final real-world check: play the most demanding game you own for a full session, since some games stress the GPU in ways synthetic benchmarks don't.

Monitor Temperatures and Fine-Tune

Throughout every step above, keep an eye on temperature — it's both a safety limit and often the real ceiling on how far you can push an overclock. Watch three numbers in your OSD:

• GPU core (edge) temperature — keep it under roughly 80-83°C for NVIDIA cards under sustained load; most cards will throttle on their own well before any damage occurs
• Hotspot/junction temperature — the hottest point on the die, typically 15-25°C above core temp; NVIDIA cards generally stay healthy under ~95°C, AMD cards are rated for junction temps up to ~110°C depending on model, but lower is always better for long-term wear
• Memory temperature — especially relevant for GDDR6X cards, which run hotter than standard GDDR6; keep it under 100-105°C where your software reports it

If temperatures run hot, your options are: increase fan curve aggressiveness (both Afterburner and Adrenalin let you set a custom fan curve tied to temperature), improve case airflow, or simply back off your power limit and voltage rather than chasing the last few MHz. A cooler, slightly lower overclock that holds its boost clock steady will usually outperform a hotter one that throttles under sustained load — check your final result against the Step 1 baseline benchmark to confirm you actually gained frame rate, not just a higher number on a slider.

Once you've settled on final values, save the profile in your overclocking software and, if it supports it, set it to auto-apply on Windows startup — otherwise you'll need to reopen the app and reapply manually after every reboot. Revisit your overclock after major driver updates or new game releases; a setting that was rock-solid for months can occasionally become unstable in a specific new title that stresses the card differently than anything in your original stress test. If that happens, drop back to your saved stock profile, confirm the new title runs fine there, and re-tune from a slightly lower starting point rather than assuming your whole overclock has gone bad.