Colored squares across your desktop. Pink flashes in a boss fight. Green pixels crawling over the Windows logo. If you’ve never seen GPU artifacting before, it’s the kind of thing that makes you wonder if you just torched a $700 card. Sometimes you did. More often, it’s a driver that didn’t uninstall cleanly, a fan choked with cat hair, or a sag-stressed PCIe contact that costs five bucks to fix. The order you check things matters. Software first, then hardware, then thermals. Skip ahead and you’ll spend a weekend chasing a problem that wasn’t there.
The quick diagnosis (30-second check)
Don’t crack the case yet. First question: are the artifacts showing up on the Windows desktop, or only inside a game? Desktop artifacts at idle usually point at driver corruption or a dying display output. Gaming-only artifacts point at thermals, VRAM under load, or voltage droop. Reboot. That’s it, just reboot. It clears a surprising number of one-off driver glitches and costs you 30 seconds.
Now watch the boot sequence carefully. If you see artifacts on the motherboard POST screen or the Windows logo (before any driver loads), you’ve got a hardware problem. If the screen is clean until Windows finishes booting and then artifacts appear, it’s almost certainly driver-side. That single observation saves hours of guessing.
Most likely cause – thermal throttling or VRAM failure
Heat kills VRAM cells. Modern GDDR6X chips are rated to roughly 95C, but they get unstable before they hit that ceiling. If your hotspot temperature is climbing past 95C under load, you’re not throttling, you’re cooking. Pull up HWInfo64 or GPU-Z, run a game for ten minutes, and watch the hotspot reading. Anything above 90C sustained is a red flag in 2026.
The fix isn’t glamorous. Clean the GPU fans. Dust mats down between the fins and turns your heatsink into an insulator. A can of compressed air and five minutes of attention drops temps noticeably. If you’ve owned the card three-plus years and you’re comfortable with the warranty implications, a repaste with a quality compound can buy you another 5-10C of headroom. Most 2026 cards ship with decent factory paste, so don’t repaste a brand-new card chasing imaginary gains.
VRAM failure looks different. It gets worse over weeks, not minutes. Memory cells fail one at a time, and the artifacts spread. Run MemTestGPU or OCCT’s VRAM check for at least an hour. If you see errors logged, the chips are dying and no amount of repasting will save you.
Second most likely cause – driver corruption or overclock instability
Driver corruption is sneaky. A Windows update that interrupts a GPU driver install, a power cut mid-update, or an old driver that didn’t fully uninstall when you swapped cards. The system runs, but it’s in an inconsistent state, and artifacts pop up at random.
The fix is Display Driver Uninstaller. Boot into Safe Mode, run DDU, choose the clean uninstall option, then reboot and pull the latest driver directly from Nvidia or AMD’s site. Not Windows Update, not GeForce Experience auto-install. The clean .exe from the manufacturer. This single procedure resolves a wild number of artifact reports on r/buildapc.
Overclock instability is the other big one. If you bumped the memory or core clock in MSI Afterburner months ago and forgot about it, a driver update can quietly destabilize what used to be a rock-solid OC. Dial it back to stock and recheck. If the artifacts vanish, you’ve got your culprit. Re-tune carefully if you want the performance back.
The weird one (rare but happens) – PCIe slot issue or PSU rail droop
This one catches people. PCIe slot contacts can develop a thin layer of oxide or pick up dust over a few years, especially in dusty rooms. The card boots fine, but at heavy 3D load when the bus is hammering data through, the connection flakes and you get artifacts. Power down, pull the GPU, and clean the gold edge connector with 99% isopropyl alcohol on a lint-free cloth. Reseat firmly. Make sure the locking tab on the PCIe slot clicks.
PSU rail droop is the other oddball. An aging or undersized PSU can’t hold 12V steady when the GPU pulls a transient spike. You’ll see clean idle behavior and clean light gaming, then artifacts only when the card pulls full power. Check with a multimeter or a proper PSU tester if you have one. If you’re running a 650W unit with a card that recommends 850W, that’s your problem. Run your build through OuterVision’s PSU calculator and upgrade if you’re under-spec.
Step-by-step fix
Here’s the diagnostic order that won’t waste your weekend:
(a) Note exactly when artifacts appear. Boot screen? Desktop only? Specific game? Write it down before you start changing things or you’ll forget.
(b) Reboot. Yes, really. One in five cases ends here.
(c) Clean driver install. Boot Safe Mode, run DDU, reboot, install the latest WHQL driver from Nvidia or AMD directly. Don’t let Windows Update touch it.
(d) Run HWInfo64 and FurMark together for five minutes. Watch the hotspot temp. Above 95C, you’ve got a thermal problem. Stop here and fix airflow first.
(e) Power down. Pull the GPU. Look at the PCIe edge connector. Dusty or discolored? Clean it with isopropyl on a cotton swab.
(f) Reseat the card firmly. Check the locking tab actually engages.
(g) If you’ve got a second PCIe x16 slot, try the card there. Slot failures are rare but they exist.
(h) Check the power cable. 12V-2×6 connectors on 40-series and 50-series cards are notorious for partial seating. Push until you hear the click. Same for 8-pin PCIe cables.
(i) If you’ve got a spare PSU lying around, swap it in. Borrowed PSUs from friends count.
(j) Run MemTestGPU overnight. Eight hours of clean memory checks means VRAM is healthy. Errors logged means the chips are failing and you’re heading toward RMA.
When it’s not fixable – what to replace
You’ve reinstalled drivers three times, repasted, reseated, swapped PSUs, and the artifacts haven’t budged. That’s a hardware failure inside the card itself, and you’re looking at replacement or RMA.
If you’re inside the warranty window (most 2026 cards ship with 3-year coverage from Nvidia’s AIB partners and AMD’s board partners), contact the manufacturer first. Don’t try Amazon returns past the 30-day window, they’ll bounce you to the manufacturer anyway. Box the card, screenshot the artifacts, save your purchase receipt.
Out of warranty is harder. Third-party repair shops can sometimes reflow BGA chips on the GPU die and bring a card back from the dead, but you’re looking at roughly $150 for the work with no guarantee. Weigh that against a current-gen mid-range card at $400-1000. Often the math points to replacement. Newegg and Amazon both run trade-in programs that’ll take your dead card for parts credit, which softens the blow a little.
A few more questions
Are artifacts reversible if I catch them early?
Sometimes, yes. If the cause is thermal stress and you fix the cooling before the VRAM cells permanently degrade, you can dodge the bullet. Same for driver corruption, that’s fully reversible with a clean DDU install. What’s not reversible is physical chip damage. Once VRAM cells have died from prolonged overheating or voltage spikes, they’re gone. The lesson: don’t ignore early artifacting hoping it’ll go away.
Will undervolting help with artifacting?
It can, if the root cause is heat or PSU droop. Undervolting drops power draw and temperatures, which gives marginal VRAM cells more headroom. MSI Afterburner’s curve editor lets you trim 50-100mV without losing much performance on most modern cards. Won’t fix a card that’s already physically failing, but it’s a smart preventive move if you’re running hot. Just don’t undervolt and then blame undervolting when something else goes wrong.
How do I know if it’s the GPU or the monitor cable?
Easy check. Swap the DisplayPort or HDMI cable for a known-good one. If artifacts vanish, it was the cable. Also try a different monitor if you’ve got one handy. Cable artifacts usually look like horizontal lines, flickering, or signal dropouts, not the colored block patterns you get from VRAM failure. If artifacts also appear on the BIOS POST screen, it’s not the cable, it’s the card.
