Graphics cards attract blame with remarkable efficiency. A game crashes, the screen turns black and attention swings towards the most expensive board in the case. Sometimes the GPU is guilty. Sometimes the culprit is a cable worth less than lunch.
This fault atlas starts with what you can actually observe. A symptom narrows the search; it does not announce a diagnosis.
No display: first establish what “nothing” means
No picture, no power and no successful boot are three different problems that happen to share a dark rectangle. Check that the monitor is powered, set to the correct input and connected to the graphics card. With the system switched off, inspect the external connection and try a known-working cable or display if one is available.
A motherboard display socket only produces a picture when the processor and platform support integrated graphics. Spinning fans prove that some electricity arrived; they do not issue a certificate saying the computer booted. Note diagnostic lights or beep codes and read the motherboard manual before translating them by instinct.
AMD’s boot troubleshooting guidance separates these starting points. Internal work should be left to somebody competent, with mains power disconnected and the correct manuals to hand. Never swap modular PSU cables merely because the plugs appear to fit; the wiring at the other end can differ.
Blocks, sparkles and the temptation to diagnose “dead VRAM”
Visual artefacts include coloured blocks, flashing polygons, lines, texture corruption and repeating patterns. Their timing matters. Do they appear before Windows loads, in one game, after ten minutes of heat, or everywhere from the first moment?
Possible causes include software, unstable clocks, memory faults, overheating and the display-signal path. A corrupted texture is evidence; it is not a tiny on-screen coroner’s report.
Capture a screenshot as well as a phone photograph or short video if possible. Corruption visible in a screenshot existed in the captured image before it reached the monitor, which helps move the investigation upstream. Corruption absent from the screenshot gives the cable, output and display more reason to be examined, although capture behaviour can complicate the result.
For example, if sparkles disappear when only the cable changes, the signal path deserves attention before the GPU is condemned. If corruption appears in saved screenshots across several applications, the investigation must go further. These are diagnostic examples, not proof of one failed part.
A black screen that comes back has told you something
Windows can attempt to recover when graphics work takes too long. The screen may go black while audio continues, then return after a few seconds. That recovery event does not by itself prove permanent hardware damage.
Record whether the picture returns, the exact message, the application, temperatures and any recent driver or settings change. If the machine is otherwise safe, compare its ordinary manufacturer-supported settings with any overclock or undervolt removed, and try another suitable output or cable.
Do not lengthen or disable TDR recovery merely to hide the symptom. Microsoft describes those registry controls as driver-development testing tools, not a domestic cure for black screens.
A game closing and a whole PC losing power are not the same event
A crash to desktop, a frozen image, a driver reset and an instant restart produce very different trails. Write down which one occurred. System RAM, CPU settings, game files, drivers, the motherboard, temperatures and power delivery can all dress up as a graphics fault.
Compare one game with several, desktop use with 3D load, and the same task before and after the most recent change. Windows logs and error codes can add evidence, but they rarely stride into the room carrying the culprit’s name.
AMD’s system-stability guidance recommends recording changes and results systematically. Controlled substitution with known-good hardware can help a competent technician; it is not an invitation to connect visibly suspect electrical hardware to another PC.
Temperature is a comparison, not a universal red line
Core, hotspot and memory temperatures are different measurements, and normal behaviour varies with model, cooler, workload, fan profile and room conditions. Comparing one sensor from one GPU with a different sensor from another is an excellent way to manufacture alarm.
More useful clues are a large change from the card’s usual behaviour, falling performance as it warms, repeated heat-related crashes or cooling that is plainly not doing its job. Dust, obstructed airflow, a failed fan, poor cooler contact or incorrect thermal work are all possibilities.
A stationary fan at idle can be deliberate. AMD documents Zero RPM operation on supported configurations; inspect the exact model before declaring fan-stop a fault. Grinding, scraping, missing blades, improvised wiring or a fan that never starts under conditions where it should are rather less philosophical.
Keep fingers and tools away from moving fans. Contact the seller of a recently purchased card before removing its cooler or changing pads. Our GPU and component testing page explains GPUsed’s process; it is not a home stress-test prescription for unsafe hardware.
Coil whine is irritating; crackling is a different conversation
Coil whine is a high-pitched electrical sound that can change with frame rate or load. ASUS explains it as vibration associated with changing electrical current. Noise alone does not establish that a GPU is failing, although a loud card can still be unpleasant in a quiet room.
If everything is otherwise safe, note whether a frame-rate cap changes the sound and whether it seems to come from the GPU or PSU. Never open a PSU to investigate. Grinding points more readily towards a fan; crackling, a burning smell or visible electrical damage belongs back in the stop-testing box.
Bent metal, loose ports and the history written on the board
Ports and brackets
A bent bracket can prevent proper seating. A loose HDMI or DisplayPort socket, cracked housing or output that works only while the cable is held at an angle may worsen with repeated use. Do not force it. Record exactly which outputs work.
Liquid and corrosion
Green or white residue, staining, darkened contacts and rust can point to contamination. Do not power a visibly wet or corroded card to “see what happens”. Dry-looking is not the same as electrically sound.
Repairs, firmware and missing parts
Replacement fans, pads, water blocks and board repairs are not automatically bad. The useful questions are why the work was done, which parts were used, whether the originals remain and what evidence exists afterwards. Modified firmware needs especially clear disclosure. Software and physical inspection can reveal inconsistencies, but neither can reconstruct an entire history from silence.
Send support a timeline, not a verdict
Include the exact GPU, CPU, motherboard, RAM and PSU; the game or application version; the exact error; when it happens; and the safe checks already attempted. Add photographs or a short recording where they reveal something words cannot.
Keep full serial numbers and order details out of public posts. Share them privately with the appropriate retailer or manufacturer. For a GPUsed purchase, contact us about the order before attempting a repair or selling the card, and read the current returns information.
Selling a GPU with a known fault
Describe the symptom and disclose repairs, missing parts and visible damage. “Untested” should mean that you genuinely do not know whether the card works, not that you know precisely what it did and would rather leave the buyer an archaeological project.
GPUsed considers suitable faulty and genuinely untested GPUs individually. If this is your own card rather than an unresolved issue with a recent purchase, send the details for a personalised quote. Never power unsafe hardware merely to make the form look more complete.