How to Overclock a GPU: 12 Steps, 60 Min [2026]

Nvidia confirmed it is skipping a new gaming GPU generation this cycle, and street prices on current cards are still elevated after the surge that pushed the RTX 5090 close to $4,500 at retail. That combination changes the math for a lot of PC gamers in 2026. If the flagship lineup you already own is going to stay the flagship lineup for a while longer, squeezing extra frames out of the silicon you have makes more sense than waiting on a replacement that isn’t coming.

This tutorial walks through how to overclock a GPU using MSI Afterburner, the free tool most reviewers and enthusiasts still reach for first. It covers Nvidia’s RTX 50-series (Blackwell) and AMD’s RX 9000-series (RDNA 4) cards, in 12 steps you can finish in about an hour, plus a longer stability pass you’ll want to let run in the background. By the end you’ll have a saved profile, a stress-tested clock offset, and a clear idea of what to do when something goes wrong.

The used and open-box market is feeling the same pressure. With fewer new cards entering retail channels this year, more buyers are shopping secondhand RTX 40-series and RTX 50-series cards, and a well-documented overclock is one of the few ways to close the gap between what you can afford and what you actually need for a demanding release. None of that requires new hardware. It requires about an hour, a free download, and the patience to test one change at a time.

·

Google · Preferred Sources

Don't miss new tech stories on Google

Add Tech Insider once in the Google app and our stories appear in your news suggestions.

Add Now

What GPU Overclocking Actually Does in 2026

Every GPU ships with a conservative clock speed and voltage curve set by the manufacturer to guarantee it works in the worst-case sample, the worst-case case airflow, and the worst-case ambient temperature. Overclocking raises the core clock, memory clock, or both, past that conservative baseline so the card runs closer to what your specific chip and cooler can actually handle. It’s the same silicon, running faster, because you’ve replaced a one-size-fits-all setting with one tuned to your hardware.

The gains are real but modest. Across the community data and reviewer testing referenced throughout this guide, a conservative overclock typically adds somewhere in the 5-10% FPS range, a moderate one lands around 10-15%, and an aggressive overclock on a chip with unusually good headroom can approach 20% in favorable titles. That’s not a free generational upgrade. It’s closer to the difference between a card’s base clock and its highest factory-overclocked variant, achieved on your own card instead of paid for at checkout.

Two separate knobs drive that gain. Core clock offset increases how fast the GPU’s shader and render pipelines execute instructions. Memory clock offset increases how quickly the card can move data in and out of VRAM, which matters more in memory-bandwidth-hungry games and at higher resolutions. RTX 50-series cards use GDDR7 memory, which has more headroom for offset increases than the GDDR6 used on RX 9000-series cards, but both benefit from tuning independently rather than assuming one setting fixes both. GDDR7’s higher native signaling rate is also why Nvidia’s cards responded so well to the memory-focused Afterburner update mentioned above. The underlying memory controller simply had more room to give before hitting a wall, where GDDR6 on RX 9000-series cards was already running closer to its practical ceiling out of the box.

None of this requires touching the BIOS or opening the case. Every step below happens in software, through MSI Afterburner’s sliders, and every change is reversible with a single click. That’s a meaningfully lower-risk project than the driver-level tuning covered in our XMP/EXPO RAM guide, where a bad profile can occasionally stop the system from posting.

Prerequisites: Hardware, Software, and Versions You Need

You don’t need exotic hardware to follow along, but a few things need to be in place before you touch a single slider. Skipping this checklist is the single most common reason overclocking sessions go sideways, and it’s a longer list than most quick-tip videos suggest because monitoring software matters just as much as the tool doing the actual tuning.

• A discrete GPU from Nvidia’s RTX 50-series (5070, 5070 Ti, 5080, 5090) or AMD’s RX 9000-series (RX 9070, RX 9070 XT). Older cards work with the same software and largely the same steps.
• MSI Afterburner 4.6.7 or newer. This build followed the long-delayed 4.6.6 release and added native RTX 50-series support plus extended GDDR7 memory overclocking headroom, alongside the unofficial RDNA 4 support that arrived in 4.6.6. Grab it from Guru3D’s official Afterburner download page, not a third-party mirror.
• RivaTuner Statistics Server (RTSS), which installs alongside Afterburner and powers the on-screen display overlay.
• The latest GeForce or AMD Software: Adrenalin Edition driver. Nvidia shipped a hotfix, build 595.76, after users on 595.71 reported an artificial voltage cap that quietly cut overclocking headroom by roughly 200 MHz. If your clocks feel unusually limited, check your driver build before you blame your silicon.
• A stress-testing tool: 3DMark, Unigine Superposition, or FurMark. Pick at least one, the walkthrough below uses all three at different stages.
• HWiNFO64 and GPU-Z for a second opinion on temperatures, power draw, and VRAM health that Afterburner’s own overlay doesn’t always surface in detail.
• Display Driver Uninstaller (DDU), kept on hand in case a driver update goes wrong mid-project.
• A case with decent airflow and, ideally, a power supply with headroom above your GPU’s rated draw. If you’re still finalizing your build, our gaming PC build guide and 12VHPWR cable installation walkthrough are worth reading first, since a marginal power connection will surface as instability that looks like a bad overclock.

Budget about 60 minutes for the core walkthrough below, plus however long you let the final stability test run. Most guides suggest a minimum of one hour for a final pass, with two to three hours being closer to what you’d want before trusting a profile for daily use.

One prerequisite that’s easy to overlook: confirm your power supply actually has headroom before you raise the power limit slider. A system running an already-marginal PSU close to its rated capacity is a worse candidate for overclocking than one with comfortable spare capacity, since you’re specifically asking the GPU to pull more power at the exact moment you’re testing for stability. If you’re not sure how much headroom you have, check your PSU’s rated wattage against your full system’s estimated draw under load before proceeding, rather than finding out mid-stress-test.

Is GPU Overclocking Safe? Warranty and Risk, Explained

Software-based overclocking through Afterburner, the Nvidia App, or AMD Adrenalin is safe in the sense that matters most: modern GPUs have hardware protections that throttle clocks, cut power, or shut the card down entirely before permanent damage occurs. You are extremely unlikely to destroy a card by moving a slider too far. The realistic risks are smaller and more annoying: driver crashes, game crashes, visual artifacts, or a system that won’t boot into Windows until you reset the profile.

Both Nvidia and AMD have long treated overclocking through standard, vendor-supported software controls as something that does not void your warranty on its own. Their stated position, reiterated by both companies for years, is that a warranty claim can be denied if overclocking causes physical damage, such as a fried VRM or a burned-out memory chip, not because you raised a clock slider. In practice, using Afterburner’s core and memory offsets, or Adrenalin’s own tuning panel, stays well inside that boundary. Where things get riskier is manual voltage modification beyond what the software exposes, which this guide does not cover and which most beginners should avoid entirely.

The more common failure mode isn’t hardware damage. It’s applying an unstable profile, closing Afterburner, and forgetting what you changed. That’s why every step below insists on testing one variable at a time and writing down what you changed before you change the next thing. An overclock you can’t reproduce or explain isn’t a stable overclock, even if it happens to work for a week.

One more caveat worth stating plainly: overclocking headroom varies by individual chip, not just by model. Two RTX 5080 cards bought on the same day can hit different ceilings before they crash, a phenomenon enthusiasts call the silicon lottery. Nothing in this guide, or any guide, can promise you a specific number. What it can do is give you a safe, repeatable process for finding your card’s actual ceiling instead of guessing.

It’s also worth distinguishing this kind of overclocking from what an AIB partner already did at the factory. A card sold with an “OC” or “Gaming” suffix in its name typically ships with a modest factory bump baked into its firmware, applied to a batch of chips the manufacturer has already validated at scale. What this guide walks through goes further than that, tuning your specific card individually rather than relying on the conservative, one-size-fits-all margin a factory overclock has to leave for every unit in the batch. Workstation and professional-visualization cards built on the same silicon are a separate case. They’re often certified for specific software and industries, and pushing them past their validated clocks can matter more for driver stability in professional applications than it does for gaming.

The Best GPU Overclocking Tools in 2026

MSI Afterburner remains the default recommendation because it works on both Nvidia and AMD cards through one interface, ships the RivaTuner overlay for free, and gets updated fast enough to track new hardware. TweakTown’s coverage of the 4.6.6 update confirmed the RDNA 4 support arrived as an unofficial, experimental layer, since MSI doesn’t manufacture AMD cards, while RTX 50-series support landed officially. The follow-up 4.6.7 release pushed further, and Tom’s Hardware reported it raised the GDDR7 memory overclocking ceiling on RTX 50-series cards enough to let some owners push VRAM transfer rates up to 36 GT/s, board and firmware permitting.

Vendor-native tools are worth running alongside Afterburner rather than instead of it. Nvidia bundles GPU tuning into the Nvidia App, which includes an automated scanning feature that tests a safe core clock curve for you in a few minutes. It’s a reasonable starting point, though it tends to land more conservative than what you can find manually with patience. AMD’s equivalent lives inside AMD Software: Adrenalin Edition, under its own tuning panel, which offers preset Auto, Custom, and Manual modes for both overclocking and undervolting RX 9000-series cards.

Tool	Type	Cost	Best for
MSI Afterburner + RTSS	Third-party OC utility + overlay	Free	Core/memory offsets, power limit, fan curves, works on Nvidia and AMD
Nvidia App tuning scanner	Vendor-native, Nvidia only	Free	Fast automated baseline core clock curve
AMD Adrenalin tuning panel	Vendor-native, AMD only	Free	Auto-tuning and undervolting presets for RX 9000-series
3DMark	Synthetic benchmark suite	Paid (free demo)	Repeatable, comparable before/after scoring
Unigine Superposition / Heaven	Synthetic stress test	Free / paid tiers	Longer-duration thermal and stability stress
FurMark	Power/thermal stress test	Free	Worst-case thermal and power-draw validation
HWiNFO64	Hardware monitoring	Free	Detailed sensor logging, VRAM and hotspot temps
GPU-Z	Hardware monitoring	Free	Quick spec, clock, and VRAM health checks

You don’t need every tool in that table running at once. Afterburner plus RTSS for the actual tuning and overlay, HWiNFO64 for a second set of temperature readings, and one benchmark from the list for before-and-after comparisons covers everything the rest of this guide assumes you have installed. RTSS’s overlay is also worth customizing beyond the defaults, since you can reposition it, change its font size for a 4K display, and set it to log every session to a file automatically, which turns every future benchmark run into a record you can compare against later without extra setup.

How to Overclock Your GPU: Step-by-Step Walkthrough

Work through these 12 steps in order. Each one builds on the last, and skipping ahead is the fastest way to end up with a crash you can’t explain.

Part 1: Baseline, Core Clock, and First Stability Pass (Steps 1-7)

1. Update your driver and clean out the old one. Install the latest GeForce or Adrenalin driver before you install Afterburner. If you’re jumping more than one driver version, run DDU in Safe Mode first to remove the old driver completely rather than installing over it. A quick way to confirm what you’re currently running:

# PowerShell: check installed GPU driver version and date
Get-CimInstance Win32_VideoController | Select-Object Name, DriverVersion, DriverDate

2. Install MSI Afterburner and RivaTuner Statistics Server. Both come bundled in one installer. Accept the RTSS install when prompted, since Afterburner’s on-screen display depends on it.
3. Set up your on-screen monitoring overlay. In Afterburner’s settings, enable the on-screen display and add core clock, memory clock, GPU temperature, hotspot temperature, and power draw to the overlay. You’ll be watching these numbers constantly for the rest of this walkthrough. If you’d rather monitor from the command line while a game or benchmark runs in a separate window, Nvidia’s own system management interface can poll the same data:

# Command Prompt: poll core clock, memory clock, temperature, and power draw every 2 seconds
nvidia-smi --query-gpu=clocks.sm,clocks.mem,temperature.gpu,power.draw --format=csv -l 2

4. Run a baseline benchmark and write down the score. Before changing anything, run your chosen benchmark, 3DMark Time Spy or Speed Way for Nvidia cards, Unigine Superposition for either vendor, at stock settings. Record the score, the average clock speed, the peak temperature, and the average power draw. This number is what every later step gets compared against, and it’s the step people skip most often.
5. Raise the power limit and temperature limit sliders first. Before touching core or memory clocks, move Afterburner’s power limit slider up, most RTX 50-series cards allow roughly a 15-RX 9000-series cards typically achieve 5–10% FPS gains from overclocking in software, not a hard cap at 10% for a 20% increase, as the claim conflates achievable gain with an impossible 20% target for this series [6]. Raise the temperature limit slider to its maximum as well. Neither change increases clock speed on its own. Both just remove the ceilings that would otherwise throttle the increases you’re about to make.
6. Increase the core clock offset in small steps. Move the core clock slider up by 10-15 MHz. Resist the urge to jump by 100 MHz to save time. Small, testable increments are what let you find your actual ceiling instead of guessing past it.
7. Apply and test each increment with a short benchmark loop. After each core clock bump, click Apply, then run a short benchmark pass, 10-15 minutes is enough at this stage. Watch for driver timeout errors, screen flicker, or a benchmark crash. If it passes clean, go back to step 6 and raise the offset another 10-15 MHz. If it fails, drop back down by 15-25 MHz below the last stable point and treat that as your practical core clock ceiling for now.

Part 2: Memory Clock, Power Limit, and Locking In Your Profile (Steps 8-12)

8. Leave core clock alone and move to memory. With your stable core offset locked in, increase the memory clock offset by 50-100 MHz per step. RTX 50-series cards on GDDR7 generally tolerate larger memory offsets than RX 9000-series cards on GDDR6, so don’t assume the same increment size works for both.
9. Test each memory increment specifically for artifacts. Memory instability rarely shows up as a crash. It shows up as flickering textures, colored specks, or corrupted geometry while the benchmark otherwise keeps running. Watch the screen itself, not just the FPS counter, during this stage. The moment you see artifacting, drop the memory offset by 100-200 MHz below that point and lock it there.
10. Recheck your power limit once both offsets are set. Higher core and memory clocks both pull more power. If you’re seeing clock speeds drop under sustained load despite passing your earlier short tests, raise the power limit slider further, up to its maximum, before assuming you’ve hit a clock ceiling. A lot of what looks like instability is actually a power limit quietly capping you.
11. Tune the fan curve. Open Afterburner’s fan curve editor and set a custom curve that ramps up earlier than the stock curve, most factory curves are tuned for acoustics first and cooling second. Aim to keep the card comfortably under your target temperature ceiling (more on specific numbers in the next section) rather than reacting to heat after the fact.
12. Run a long-form stability test, then save and auto-apply the profile. Once every short test passes clean, run a stress test for at least 60 minutes, ideally 2-3 hours, using FurMark or an extended Unigine loop. If it survives clean, save the configuration to one of Afterburner’s profile slots and enable “Apply overclocking at system startup” in the settings so it persists across reboots.

Once you’ve got a profile you trust, Afterburner also supports loading specific profiles from the command line, which is useful if you want a quick way to revert to a safe baseline without opening the GUI, for example, right before a driver update:

:: Load saved profile 1 and exit Afterburner's window (adjust install path as needed)
"C:Program Files (x86)MSI AfterburnerMSIAfterburner.exe" -Profile1 -Q

That same command works well dropped into a desktop shortcut or a Windows Task Scheduler entry set to run at login, giving you a one-click way back to your tuned profile after any change that resets GPU settings, like a clean driver reinstall. Creating that scheduled task from the command line, rather than clicking through the Task Scheduler GUI, takes one line:

:: Create a scheduled task that loads your saved Afterburner profile at every logon
schtasks /create /tn "LoadGPUProfile" /tr ""C:Program Files (x86)MSI AfterburnerMSIAfterburner.exe" -Profile1 -Q" /sc onlogon

Run that once from an elevated Command Prompt and the profile reapplies itself automatically every time you log in, independent of whether Afterburner’s own “apply at startup” setting survives a future driver update.

Recommended Starting Points by GPU Generation

These are starting points to test from, not guaranteed results. Every chip is different, and the only way to know what yours can hold is to work through the step-by-step process above. Treat the table as where to begin your first round of testing, not where to stop.

Reference and Founders Edition cards are a reasonable baseline for these numbers, but AIB (add-in-board) partner cards vary considerably from each other even within the same GPU model. A triple-fan, factory-overclocked flagship card from one of the larger AIB partners typically has more thermal headroom than a compact dual-fan card aimed at small-form-factor builds, simply because it has more heatsink surface area and airflow to work with. If you’re on a compact card, start at the low end of these ranges and pay closer attention to the temperature guidelines in the next section before pushing further.

GPU family	Starting core offset	Starting memory offset	Power limit	Notes
RTX 5090 / 5080	+25 to +50 MHz	+500 MHz	+15%	GDDR7 headroom is generous, some reviewers have pushed VRAM transfer rates toward the mid-30 GT/s range on select cards
RTX 5070 Ti / 5070	+25 to +50 MHz	+400 MHz	+15%	Similar approach to the 5080/5090, generally slightly less thermal headroom on cooler-constrained AIB models
RX 9070 XT / 9070	+25 to +50 MHz	+300 MHz	+10%	RDNA 4 support in Afterburner is still unofficial, cross-check readings against AMD Adrenalin’s own tuning panel

Reviewer testing gives a sense of the ceiling on a good sample. TheFPSReview’s overclocking pass on the RTX 5090 Founders Edition pushed the card to a final boosted core clock around 2968 MHz with memory around 31 Gbps in their testing, on a card cooled well enough to sustain it. That’s closer to the high end of what’s achievable than a realistic starting expectation, but it’s a useful reference for what well-cooled Blackwell silicon can do once you’ve worked through the increments yourself. If you’re deciding between the RTX 5070 Ti and RX 9070 XT in the first place, our side-by-side comparison covers how the two cards stack up before overclocking even enters the picture.

Temperature and Power Limit Guidelines

Clock speed gets all the attention, but temperature is what actually determines whether an overclock is safe to run every day. Modern GPUs throttle automatically well before hitting a damaging temperature, so the guidelines below are about sustained performance and long-term comfort, not a hard safety cliff.

Core temperature zone	What it means	Recommended action
Under 80°C	Comfortable, sustainable for 24/7 use	Good target for a daily-driver overclock
80-85°C	Acceptable under sustained full load	Fine for gaming sessions, keep an eye on fan noise and case airflow
85-88°C	Getting warm, approaching where some cards begin throttling	Improve the fan curve or case airflow before adding more clock speed
Above 88-90°C	Throttling likely, and headroom for further overclocking is gone	Back off core or power limit and address cooling first

Watch VRAM and hotspot temperatures separately from core temperature, especially on memory-heavy overclocks. HWiNFO64 breaks these out individually where Afterburner’s default overlay sometimes only shows the core figure. A card that looks fine on core temperature can still be running its memory modules hotter than is comfortable for the long term, particularly on blower-style or compact AIB coolers.

Room temperature shifts these numbers more than people expect. A profile that stays comfortably at 78°C in an air-conditioned room in January can run several degrees hotter in a closed bedroom in July with no air conditioning. If you tuned your overclock during a cooler month, it’s worth reverifying the fan curve and temperature ceiling once the seasons change, rather than assuming a profile validated once is validated forever.

How to Stress-Test and Validate Your Overclock

A profile that survives a 10-minute benchmark loop is not the same as a profile you can trust for a full evening of gaming. Stability testing happens in layers, and each layer catches a different kind of failure.

• Per-increment testing (10-15 minutes): what you use while dialing in each core and memory step, described in the walkthrough above. Fast enough to iterate, not long enough to catch everything.
• Synthetic benchmark validation (20-30 minutes): a full run of 3DMark Time Spy, Speed Way, or Unigine Superposition, ideally three consecutive runs back to back. If the score varies wildly between runs, or a later run crashes when an earlier one didn’t, the profile isn’t stable yet.
• Real-game validation (30-60 minutes): synthetic benchmarks don’t reproduce every workload. Play the most demanding game in your library for a real session. Some instability only shows up under a game engine’s specific memory access patterns, not a benchmark’s.
• Extended thermal and power stress (1-3 hours): FurMark or a long Unigine loop, run overnight or during a workday if possible, to confirm the card holds its clocks and temperatures once everything in the room has warmed up too.

Log the session instead of just watching it. A simple logging loop with nvidia-smi captures exactly when, and at what temperature, anything starts to drift:

# Command Prompt: log clocks, temp, and power draw to a CSV file every 5 seconds for later review
nvidia-smi --query-gpu=timestamp,clocks.sm,clocks.mem,temperature.gpu,power.draw --format=csv -l 5 > overclock_log.csv

If the benchmark score from your validation run doesn’t clearly beat your stock baseline from step 4, something is off, either the offsets didn’t actually apply, or you’re being throttled by power or temperature limits rather than by the clock offsets themselves. A clean overclock should show a measurable, repeatable improvement, not a coin flip.

Reading a CSV log after the fact is more informative than watching the overlay in real time, because patterns that are easy to miss live jump out once they’re on a chart. A clock speed that dips at a consistent interval usually points to power limit throttling rather than random instability. A temperature that climbs steadily for the first 20 minutes and then plateaus is normal thermal saturation. A temperature that never plateaus and keeps climbing for the full duration of a long test is a cooling problem worth solving before you add any more clock speed on top of it.

Common GPU Overclocking Mistakes to Avoid

Most bad overclocking experiences trace back to one of these habits rather than to genuinely bad silicon.

• Skipping the baseline benchmark. Without a stock score to compare against, you have no way to confirm your overclock is actually doing anything, or worse, that it’s quietly making things slower through thermal throttling.
• Raising core and memory clocks at the same time. When something crashes, you won’t know which change caused it. Tune one variable fully before touching the other, every time.
• Jumping in large increments to save time. A 100 MHz jump might work, or it might crash at a point well below what smaller, tested steps would have found. Large jumps trade precision for speed and usually cost more time overall once you factor in troubleshooting.
• Watching only core temperature. VRAM and hotspot temperatures can run meaningfully hotter than the core figure most overlays show by default, especially under memory-heavy overclocks.
• Testing only with synthetic benchmarks. Some instability is workload-specific and only appears in certain games, not in 3DMark or Unigine. Always validate with at least one real game before trusting a profile.
• Forgetting to enable auto-apply on startup. A great profile that resets every time Windows boots isn’t doing anything for you day to day. Save it, and confirm it’s set to apply automatically.
• Overclocking on an outdated or buggy driver. The 595.71 voltage-cap issue mentioned earlier is a good example: hardware and software both need to be current before you start chasing a clock ceiling that might just be a driver bug.
• Trusting a single short pass and calling it done. A profile that survives one 10-minute loop can still fail on a longer session once the case has fully warmed up. Treat the short tests as a filter for obviously bad settings, not as final proof of stability.
• Dismissing new coil whine as irrelevant. A faint, high-pitched noise that appears only under an aggressive overclock is usually harmless, but a sudden, loud increase in coil whine can be an early signal that you’ve pushed power delivery further than that particular card’s components are comfortable with. If it’s new and noticeably louder than before, back off a step and see if it goes away.

Troubleshooting: Fixing Crashes, Artifacts, and Black Screens

Something will eventually go wrong during this process. That’s normal, and it’s exactly why the step-by-step approach tests in small increments rather than jumping straight to an ambitious target. Knowing what each failure mode typically means saves far more time than guessing and re-testing from scratch. Here’s how to diagnose and fix the most common failures.

• Driver crashes or a TDR (Timeout Detection and Recovery) error mid-benchmark. The screen goes black for a second, then Windows recovers with a notification that the display driver crashed. Lower your core clock offset by 10-25 MHz and retest.
• Black screen on boot after applying a new profile. Restart into Safe Mode, open Afterburner, and reset the profile to default, or use DDU if the driver itself seems damaged. This is exactly why testing in small steps matters: you’ll know the last change you made before this happened.
• Artifacts or corrupted textures during a benchmark or game. This points to memory clock, not core clock, almost every time. Back off the memory offset by 100-200 MHz before touching anything else.
• System won’t wake properly from sleep after enabling auto-apply. Some combinations of profile and sleep state don’t play well together. Disable “apply at startup,” and instead apply the profile manually with the Afterburner shortcut described earlier, right after waking the system.
• Benchmark score is lower than your stock baseline. This usually means thermal or power throttling is undoing your clock offset gains. Check your temperature and power draw logs from the stress-testing step. If either one is pinned at its limit, raise the power limit slider or improve cooling before adding more core clock.
• Fan curve seems stuck or unresponsive. Running two tools that both try to control fan speed, say, Afterburner and a vendor app simultaneously, causes conflicts. Pick one tool to own fan control and disable that function in the other.
• Afterburner doesn’t fully detect or control an RX 9000-series card. Confirm you’re on the latest Afterburner build and the latest Adrenalin driver. Since RDNA 4 support is unofficial, some sensor readouts may be incomplete even when core functions work, which is where cross-checking with Adrenalin’s own panel and GPU-Z helps.
• A profile that passed every test starts crashing days later. Marginal overclocks can pass a short validation and still fail once ambient room temperature rises or dust builds up in the cooler. Back the profile off by one more increment than your last passing test to build in a small safety margin for real-world conditions.
• Overclock silently resets after a Windows update. Windows updates occasionally reset GPU driver settings. Reopen Afterburner, confirm your profile is still saved, and reapply it manually if the automatic startup option didn’t survive the update. The scheduled-task approach described earlier in the walkthrough is more resilient to this than Afterburner’s own startup toggle, since it re-triggers on every logon regardless of what the update changed.
• Two monitoring tools report different temperatures for the same card. A small gap between Afterburner’s overlay and HWiNFO64’s reading is normal and usually comes down to sampling frequency and which specific sensor each tool prioritizes. A large, consistent gap of 10°C or more is worth investigating further, since it can indicate one tool is reading a stale cached value rather than a live sensor.

Advanced Tips: Curve Editing, Silicon Lottery, and Per-Game Profiles

Once the basic flat-offset overclock described above is stable, a few more advanced techniques can squeeze out additional performance or efficiency.

Afterburner’s voltage/frequency curve editor, accessible from the main toolbar, lets you set a different clock offset at each voltage point instead of one flat offset across the board. This matters because GPU boost behavior isn’t linear. A chip might comfortably hold a higher clock at a mid-range voltage point but need a lower offset at the very top of the curve to stay stable. Curve editing takes longer to dial in than a flat offset, but it typically extracts a bit more performance for the same temperature and power budget.

Accept that the silicon lottery is real and out of your control. Two identical cards, bought from the same retailer on the same day, can land on meaningfully different clock ceilings before they crash. If your results land below the starting points in the table earlier in this guide, that’s not necessarily a sign you did something wrong. It may just be a more average chip. Chasing a number you read in a forum post, rather than the ceiling your specific card actually has, is a common source of frustration.

Afterburner supports multiple saved profiles, and it’s worth using more than one. A quiet, lower-power profile for less demanding games or media playback, and a full overclock profile for the games that actually benefit from the extra headroom, gives you a better noise and thermal experience day to day than running one aggressive profile all the time regardless of workload.

Memory overclocking has a deeper layer beyond the flat offset most people stop at: sub-timings, sometimes called memory straps. These control the exact latency between individual memory operations rather than just the overall clock speed, and tightening them can improve performance at the same clock offset, or let a card hold a higher offset than it otherwise could. This is a genuinely advanced technique, closer in spirit to manual RAM timing tuning than to anything covered in the step-by-step section above, and it’s not something Afterburner exposes directly on most cards. Treat it as a next step once flat core and memory offsets stop yielding gains, not as a starting point.

GPU Overclocking vs. Undervolting: Which Should You Choose?

Overclocking isn’t the only way to change how your GPU performs, and for a growing number of users, it isn’t even the first thing they reach for. Undervolting reduces the voltage supplied at a given clock speed rather than raising the clock speed itself, which can lower temperatures and fan noise while keeping performance close to, or in some cases matching, stock levels because the card avoids some of the throttling that comes from running hotter.

The two techniques aren’t mutually exclusive. A common approach among enthusiasts is to undervolt first, establishing a lower, cooler, more efficient voltage curve, and then apply a modest core clock offset on top of that curve once temperatures have dropped. This can produce a card that runs both faster and cooler than its out-of-box settings, though it takes more patience than either technique alone. Dedicated one-click undervolting tools built specifically for RTX 50-series cards have also started showing up in the community, which is worth exploring once you’re comfortable with the basic curve editor.

If you’re deciding where to start, pick based on what’s actually bothering you. If your card runs hot or loud under load, start with undervolting. If temperatures and noise are already fine and you specifically want more frames, start with the core and memory overclock steps in this guide. If you want both, do them in that order, not at the same time.

The methodology is nearly identical either way, which is part of why it’s worth learning the process in this guide even if undervolting ends up being your primary goal. You’re still working through Afterburner’s curve editor, still testing one increment at a time, and still validating with the same benchmarks and stress tests described above. The only real difference is which direction you’re moving the slider and what you’re optimizing for.

GPU Overclocking on Laptops and Handhelds

Everything above assumes a desktop card with its own dedicated cooler and an open power budget. Laptop GPUs and handheld gaming PCs play by different rules, and it’s worth understanding why before you try applying the same offsets.

Mobile GPUs share a thermal budget with the CPU inside a much smaller chassis, and many laptop vendors lock power limits below what the silicon itself could otherwise handle. Afterburner still works on most gaming laptops, but the realistic headroom is smaller, and temperature limits matter more because there’s simply less room for hot air to go. Undervolting tends to pay off more on laptops than a straight core clock overclock does, since the efficiency gain directly reduces the heat the smaller chassis has to dissipate.

Handhelds add a further constraint: battery life. An overclock that meaningfully raises power draw will noticeably shorten a handheld’s runtime away from a wall outlet, which is a tradeoff desktop users don’t have to think about. If you’re weighing a docked, eGPU-driven setup instead, where thermal and power limits loosen back up closer to desktop territory, our OCuLink vs USB4 eGPU comparison covers how much headroom each connection standard actually leaves on the table.

Afterburner detects most gaming laptops correctly, but a subset of vendors lock power and voltage controls at the firmware level regardless of what the software allows, as a way of protecting a thinner chassis from sustained high draw. If you install Afterburner and the power limit or core voltage sliders are greyed out or refuse to move, that’s very likely the reason, not a bug in the software. Undervolting through the curve editor tends to work even on locked-down laptops more often than raising the power limit does, since it’s reducing demand on the power delivery system rather than asking it for more.

A Complete Worked Example: Overclocking an RTX 5080 From Scratch

Here’s how the full process plays out end to end, using a typical RTX 5080 session as a worked example. Treat the specific numbers as illustrative of the process, not a guarantee for your own card. The point of walking through it start to finish is to show what a realistic session actually looks like, including the failed attempts, rather than just the final settings someone posts in a forum thread without the testing that got them there.

1. Baseline: stock settings, Unigine Superposition 1080p Extreme preset, score recorded, average core clock and temperature logged.
2. Power and temperature limits: power limit raised to +15%, temperature limit raised to maximum. No score change yet, as expected, since no clock offset has been applied.
3. Core clock, round 1-4: offset raised in four rounds of +15 MHz each, each tested with a 10-minute benchmark loop. All four rounds pass clean.
4. Core clock, round 5: offset raised another +15 MHz. Benchmark crashes with a driver timeout after roughly six minutes.
5. Core clock, final: offset dropped back to the round 4 value minus 15 MHz as a safety margin, landing at a total core offset in the +45 to +60 MHz range. Retested clean for 15 minutes.
6. Memory clock, round 1-3: offset raised in three rounds of +100 MHz, each tested with a 15-minute loop while specifically watching for on-screen artifacts. All three pass.
7. Memory clock, round 4: offset raised another +100 MHz. Faint texture flickering appears roughly ten minutes in.
8. Memory clock, final: offset dropped back to round 3’s value, roughly +300 MHz total, and retested clean.
9. Fan curve: adjusted to ramp up earlier, holding the card in the 78-82°C range under sustained load instead of drifting toward 87-88°C on the stock curve.
10. Final validation: three consecutive Unigine Superposition runs, all within a few points of each other, followed by a two-hour FurMark session with no crashes and stable clocks throughout.
11. Real-game check: a 45-minute session in the most demanding game in the library, no crashes, no artifacts.
12. Result: final benchmark score landed in the roughly 8-12% range above the stock baseline from step 1, consistent with the moderate-overclock range referenced earlier in this guide, and the profile was saved and set to auto-apply at startup.

Notice that roughly half of those steps involve backing off from a value that failed. That’s not a sign of a bad process. Finding the point just past where something breaks, and then stepping back from it, is how a stable overclock gets built, on an RTX 5080 or on any other card covered in this guide.

Frequently Asked Questions

Is it still worth overclocking a GPU in 2026?

Yes, arguably more than in previous years. With Nvidia skipping a new gaming GPU generation this cycle and GPU prices still elevated, the card you already own is likely to stay your card for longer than usual, and a free 5-15% performance gain from software tuning is a reasonable way to extend its useful life.

Does overclocking a GPU void the warranty?

Not on its own. Both Nvidia and AMD have long maintained that overclocking through standard, vendor-supported software doesn’t void a warranty by itself. A claim can be denied if the overclock causes physical damage, but simply raising a clock offset in Afterburner or a vendor tuning panel stays within normal use.

How much FPS can I actually gain from overclocking?

Expect roughly 5-10% from a conservative overclock, 10-15% from a moderate one, and up to about 20% in favorable games if your specific chip has unusually good headroom. Your mileage depends heavily on the individual card, the game, and how much thermal and power headroom your cooling and power supply leave available. Games that are more memory-bandwidth bound tend to respond better to the memory clock offset specifically, while shader-heavy workloads respond more to core clock.

Is MSI Afterburner free, and does it support RTX 50-series and RX 9000-series cards?

Yes, Afterburner is free. It offers official support for RTX 50-series cards as of the 4.6.6 and 4.6.7 releases, and unofficial, experimental support for RX 9000-series cards, since MSI doesn’t manufacture AMD hardware. For AMD cards, cross-check readings against AMD’s own Adrenalin tuning panel.

What’s safer: overclocking or undervolting?

Both are safe when done through standard vendor-supported software, since hardware protections prevent permanent damage from either. Undervolting carries slightly lower practical risk of instability since you’re reducing rather than increasing demand on the card, which is part of why it’s often recommended as the place to start.

How long should I stress-test a new overclock before trusting it?

Use short 10-15 minute tests while dialing in each individual increment, then finish with a longer validation pass of at least an hour, ideally two to three hours, plus a real gaming session, before treating the profile as safe for daily use.

Can overclocking damage my GPU?

It’s unlikely with software-based overclocking through Afterburner or vendor tools, since modern GPUs throttle or shut down before reaching damaging conditions. The realistic downsides are crashes, artifacts, or an unbootable display, all of which are fixed by resetting the profile, not by replacing hardware. Risk increases significantly if you move beyond software-exposed voltage controls into hardware-level modification, which isn’t covered in this guide and isn’t necessary to get the performance gains described here.

Do laptop GPUs overclock the same way as desktop GPUs?

The software and general process are the same, but laptops have less thermal and power headroom because the GPU shares a smaller chassis with the CPU. Realistic gains are usually smaller, and undervolting tends to be the more effective starting point on a laptop than a straight core clock overclock.

Related Coverage

• RTX 5070 Ti vs RX 9070 XT: The Quick Answer
• Gaming GPU Prices Surge: RTX 5090 Nears $4,500 [2026]
• Nvidia Skips New Gaming GPUs, Breaks 30-Year Streak [2026]
• How to Enable XMP/EXPO RAM: 12 Steps, 40 Min [2026]
• 12VHPWR GPU Cable Setup: 10 Steps, 30 Min [2026]
• How to Build a Gaming PC in 12 Steps, 90 Min [2026]
• OCuLink vs USB4 eGPU Docks: 149 vs 60 FPS [2026]