guide · static export
Hardware

Diagnose GPU Artifacts Before You Replace the Card

Checkered squares, sparkles and crashes under load are not always a dead card. Four tests separate a driver problem from a hardware defect.

by@guidox updated:10/31/26 time:45 min level:Intermediate tested on:Windows 11 24H2 works on:Windows 10, Windows 11 tools:GPU stress test
TL;DR

Reproduce the artifacts under a **known workload**, then test at **stock clocks**, with a **clean driver install** and at **lower memory clocks**. Artifacts that follow the memory clock are hardware.

Before you start

Requirements

  • A GPU stress test
  • A spare PCIe slot or a second PC
  • Thirty minutes
Step by step

Instructions

01

Capture the symptom properly

Note when the artifacts appear: idle, in one game, only after an hour, or in the desktop too. Artifacts on the desktop and in the BIOS screen point at hardware, one game points at software.

02

Remove every overclock

Reset the vendor tool and the driver panel to stock, including the memory offsets. A memory overclock that was stable in winter fails in summer, and the artifact looks exactly like a dying card.

⚠ warningFactory overclocked cards ship with a second BIOS. Check which one is active before you conclude anything.
03

Install the driver cleanly

Remove the old driver with the vendor's clean install option or a driver removal tool, then install the previous stable version. A corrupted shader cache produces artifacts that survive a reinstall.

Device Manager > Display adapters > Uninstall device
tipClear the shader cache folder after the reinstall; leftover compiled shaders are a common source of blocky flicker.
04

Stress the memory, not only the core

Run a test that hammers the VRAM for twenty minutes and watch for the first artifact. Core errors usually crash, memory errors usually draw. The distinction tells you which part is failing.

05

Check power and temperature

Read the GPU temperature, the hotspot and the power draw during the test. A hotspot above 105 degrees or a card that drops its clocks hard points at cooling or at the power path, not at the memory chips.

06

Test in another machine or slot

Move the card to a second system or a different slot. Artifacts that follow the card are the card; artifacts that stay with the machine are the board, the slot or the power supply.

Troubleshooting

Common issues

Artifacts appear only in one game

Usually the game's renderer or a mod. Verify the files and disable overlays before suspecting the card.

The screen goes black under load but recovers

A power or driver reset. Check the PCIe power cables and the PSU headroom, then the driver.

Everything looks fine in safe mode

The driver or an overlay is involved. Safe mode uses the basic display driver, which skips the GPU's compute path.

Artifacts survive a clean reinstall

Hardware, most likely VRAM. Document the test and start the warranty process with the evidence.

FAQ

Frequently asked

Can artifacts be repaired?

Not by software once they follow the card. Reflowing is a temporary fix with a poor success rate.

Does a repaste help?

Only if the hotspot is the cause. Check the hotspot temperature before you open anything.

How to undo this

Restore the overclock profile and the driver version you had before. No file on the card is changed by these tests.

Changelog · 1 entry
  1. 2026-10-31
    First version
Interactive version: checklist, feedback, related guides →

Free · no account needed · tracks your progress