Signs Your Graphics Card Is Dying: Artifacts, Crashes and What NJ Gamers Bring Me
Games crash, the screen fills with colored garbage, or the tower goes black under load with the fans still spinning. Before you price a new card, this page separates the symptoms that point at the GPU from the ones that are really the power supply or a driver. I do custom PC repair and upgrades from a drop-off shop in Somerville, NJ.
In this guide
- The symptom checklist
- Four things that look like a dying GPU and aren't
- Home tests before you spend a dollar
- What I do on the bench that you can't
- Repairable or not
- The replacement card and your power supply
- What I see on the bench in New Jersey
- Directions to Dave's Computers
- Frequently asked questions
- Sources
The Symptom Checklist
Each symptom, what it usually means, and how confident I am that the card is the problem.
A graphics card rarely dies all at once. It gets flaky under load, then at idle, then it stops posting. This is the checklist I run when a tower lands on my counter.
| Symptom | What's usually happening | Points at the GPU? |
|---|---|---|
| Artifacts: colored lines, checkerboards, blocks of wrong pixels | Bad video memory, a cracked solder joint, or an overheating core. Artifacts on the BIOS screen are worse than artifacts only in games. | Very likely |
| "Display driver stopped responding" or a game crash citing the driver | Windows reset the driver because the GPU didn't answer in about two seconds. Bad driver, unstable power, or failing chip. | Maybe, needs testing |
| Black screen under load, fans still spinning | The card dropped its output. Often power: a sagging PSU, a poorly seated cable, or failed regulators on the card. | Split between GPU and PSU |
| Fan not spinning, grinding or rattling | A worn fan bearing. The card overheats and throttles. | Yes, but it's a fan, not a chip |
| Textures flickering or loading as flat colors | Video memory errors, or a driver mismatch. Memory errors get worse as the card heats up. | Likely if it survives a clean driver reinstall |
| "Tearing" that isn't tearing (lines don't move with the frame, V-Sync does nothing) | A failing display output, a bad cable, or the monitor. | Possible, test cable and monitor first |
| PC only boots with the card removed | The card is shorting or failing its power-on check. Occasionally the slot or PSU cable. | Very likely |
| Coil whine (buzz at high frame rates) | Power inductors vibrating. Normal for many cards. | No. Not a failure sign. |
Artifacts are the one symptom I take at face value. Everything else has a second suspect.
Four Things That Look Like a Dying GPU and Aren't
Every one of these has come across my counter labeled "bad graphics card."
A power supply that can't keep up. A modern card spikes well above its average draw for milliseconds at a time. An older or undersized supply sags, and you get a black screen or hard restart that looks exactly like GPU failure. The tell: it only happens in the heaviest scenes, and it never produces artifacts.
Overheating. Dried paste, dust-packed fins and dead airflow push a healthy card to its thermal limit, where it throttles and eventually crashes. Tom's Hardware's guidance is to keep the core well below 90C; clocks that fall 100 MHz or more over a session mean throttling. That's a cleaning or a repaste, not a replacement.
A bad cable or monitor. A DisplayPort cable with a damaged pin gives you flicker, sparkles, or a signal that drops when the card ramps up. I've had customers ready to buy a card whose only problem was a cable bent behind a desk for five years.
Driver problems. A bad update or leftovers from a previous card produce crashes and texture glitches. Microsoft's documentation on Timeout Detection and Recovery is clear that "display driver stopped responding" fires whenever the GPU misses its two-second deadline, for any reason. I cover the NVIDIA version in VIDEO_TDR_FAILURE and nvlddmkm errors.
Home Tests Before You Spend a Dollar
Six checks you can do at your desk, in the order I'd do them.
Clean driver reinstall
Uninstall the driver completely, then install the current version fresh. AMD publishes the AMD Cleanup Utility for stripping out old drivers; on the NVIDIA side, Display Driver Uninstaller from Safe Mode does the same job.
Swap the cable and the monitor
Try a different DisplayPort or HDMI cable, a different port on the card, then a different monitor. Artifacts that change with the cable were never the card.
Run on integrated graphics
If your CPU has integrated graphics (most Intel Core chips without an F, and Ryzen 7000/9000 desktop parts), pull the card, plug the monitor into the motherboard, and use the PC for a day. Stable on integrated narrows it to the card or its power feed.
Watch temperatures and clocks under load
NVIDIA's app and AMD's Adrenalin software show core and memory temperature and clock speed. Run a game for twenty minutes. A core in the high 80s, or clocks sliding downward, means a cooling problem first.
Underclock it
Drop the core and memory clocks 100-200 MHz and the power limit to 80 percent. A card that crashed at stock and runs stable when asked for less has degrading silicon or power stages. A diagnostic, not a fix.
Check the power connector seating
Power down and look at the plug. On an 8-pin, press until the latch clicks. On a 12V-2x6 or 12VHPWR 16-pin, push until there's no gap between plug and socket, and don't let the cable bend sharply near the plug. Tom's Hardware's write-up on why these connectors melt comes down to fully inserting both ends and keeping strain off the plug.
What I Do on the Bench That You Can't
The steps that turn "probably the card" into "definitely the card."
The home tests narrow it down. The bench tests settle it, because I have a known-good power supply, a second test system, and the time to log a card under load.
First the card goes on my bench supply. If a card that black-screened at home runs an hour of sustained load on a supply I trust, the customer's PSU goes on the table. Then a real stress test. Tom's Hardware's graphics card testing guide covers the same toolkit I use: 3DMark or FurMark to push power and heat to the ceiling, HWiNFO or GPU-Z to log temperatures, clocks and power draw, and a game benchmark, because some faults only show with real rendering work.
Artifacts that appear as memory temperature climbs point at the memory chips. Clocks that collapse while the core is cool point at power delivery on the card. A hard crash at a consistent power level on a good supply points at the core.
When it's still unclear, the card goes into a second system. A card that fails in two machines on two supplies has run out of excuses. A card that's fine on my bench but fails in the customer's tower sends me back to that tower: slot, board power delivery, airflow. That's desktop repair work, and it's why I ask for the whole tower.
Repairable or Not
What I can fix on a graphics card, and when I'll tell you to stop spending money on it.
Repaste and thermal pad replacement. Factory paste dries out over three to five years, and the pads between the memory chips and the cooler crack. A card that runs hot and artifacts only when warm is a good candidate. I pull the cooler, clean the die, apply new paste and cut new pads. This is the most common card repair I do that ends with the customer keeping their card.
Fan replacement. A grinding or dead fan is a mechanical part, and mechanical parts get replaced. The chip is usually fine if the customer stopped gaming when the noise started.
Melted connector. Damage on the cable plug with a clean socket is a cable and usually a power supply conversation. Melted plastic or a discolored pin in the socket on the card makes it a warranty case or a replacement, because a resoldered socket carrying up to 600 watts isn't something I'll put my name on.
When replacement is the answer. Artifacts at the BIOS screen. Memory errors that survive a repaste. A card that won't post in two systems. Power stages that fail on a known-good supply. Board-level repair on a consumer card rarely makes sense against the price of a working one. I'll say so after the diagnostic, and the diagnostic fee is credited toward the graphics card replacement. I supply the card with a manufacturer warranty and test it in your tower before pickup.
The Replacement Card and Your Power Supply
The part of the quote people don't expect.
A card bought in 2019 and one bought in 2026 don't ask the same thing of a power supply. NVIDIA's own buying guidance for current mid-range cards says to check your existing supply's rating against the card's requirement, and notes that some cards ship with an adapter for the 12-pin (12V-2x6) connector instead of the older 8-pin. Upper-tier cards use that 16-pin connector as standard, and an ATX 3.x supply with a native 12V-2x6 cable is the clean way to feed one.
So when I quote a replacement card, I check the existing supply's wattage against the new card's draw, its age, and whether it has the right connector without stacking adapters. If any of those fails, the quote includes a power supply upgrade with the card. My PSU wattage guide walks through the numbers.
And if the old card died from a power problem, I want to know the supply didn't cause it. A supply that killed one card can kill the next. If the tower shows nothing at all now, not even the BIOS, that's a different tree: see computer turns on but no display.
What I See on the Bench in New Jersey
Composite examples, built from the kinds of jobs that come across my Somerville counter.
A Bridgewater customer's four-year-old gaming tower. Green checkerboard artifacts ten minutes into any game. Memory temperature was hitting 104C under load; the pads had crumbled. Repaste and new pads brought memory temps to the low 80s and the artifacts didn't come back.
A Piscataway customer's prebuilt with an upper-tier card. Black screen under load, fans spinning. The customer had already swapped in a second card, which did the same thing. On my bench supply both cards ran an hour clean. The original supply was sagging on the 12V rail under transient load. Power supply replaced, both cards fine.
A Franklin Township customer's tower that only posted with the card removed. Same halt in my second system, and visible discoloration on a power stage component. That one was a replacement, with the diagnostic credited toward the new card.
If your card is artifacting, crashing, or going black under load, bring the tower to Somerville. The $75 diagnostic ($149 for custom-built and boutique PCs) is credited toward the repair or the replacement when you approve the work.
📞 Call (908) 428-9558Drop-off shop at 75 N Bridge St, Somerville NJ · Mon-Fri 10-5, Sat 9-2
Directions to Dave's Computers in Somerville, NJ
Downtown Somerville, one block off Main Street, behind Bank of America.
From Bridgewater
Route 22 or Route 202/206 to the Somerville Circle, then Route 28 east (West Main Street) into downtown. Left onto N Bridge St.
Open in Google MapsFrom Hillsborough
Route 206 north into Somerville, right onto West Main Street, then left onto N Bridge St.
Open in Google MapsFrom Bound Brook
Route 28 west straight into downtown Somerville. Right onto N Bridge St.
Open in Google MapsFrom Piscataway
I-287 north to Exit 17 (Route 202/206 south), then Route 28 east into downtown Somerville.
Open in Google MapsFrom Flemington
Route 202 north to the Somerville Circle, then Route 28 east into downtown. Left onto N Bridge St.
Open in Google MapsFrom Princeton
Route 206 north through Montgomery and Hillsborough into Somerville. Right onto West Main Street, then left onto N Bridge St.
Open in Google MapsTimes are typical drives outside rush hour. Call 908-428-9558 before a long drive if you want to confirm I'm at the counter.
Graphics Card Failure Questions From NJ Customers
The ones I answer at the counter every week.
What are the signs a graphics card is dying?
The strongest signs are artifacts on screen (colored lines, checkerboards, blocks of garbage pixels), games crashing with a display driver error, a black screen under load while the fans keep spinning, and a fan that grinds or stops. One alone can be a driver or power problem. Two or more together, getting worse over weeks, usually means the card.
Can a bad graphics card be repaired, or does it always need replacing?
It depends on what failed. A card that overheats because the paste and pads have dried out can be repasted. A dead fan can be swapped. A melted connector on the cable side is usually a cable and power supply fix. A GPU die or memory chip failure means replacement, and I'll tell you that after the diagnostic.
How do I test if my GPU is bad without buying anything?
Swap the monitor cable, try a second monitor, do a clean driver reinstall, and run the PC on integrated graphics if your CPU has them. Watch temperatures in the card's own software while a game runs. If the crashes stop when the card is underclocked or out of the machine, the card is the suspect.
Is coil whine a sign my graphics card is failing?
No. Coil whine is a high-pitched buzz from the power components on the card vibrating at high frame rates. It's annoying and varies from card to card, but it doesn't damage anything or predict a failure. I don't treat it as a fault unless it started suddenly along with crashes or artifacts.
Could my power supply be causing what looks like a dying graphics card?
Yes, and it's one of the most common things I find. A supply that can't hold voltage under a gaming load causes black screens, restarts and driver crashes that look exactly like GPU failure. Artifacts are the exception: garbage pixels almost always come from the card. I test with a known-good supply before I blame the card.
Will a new graphics card need a new power supply too?
Sometimes. Current cards draw more power and some need a 12V-2x6 connector that older supplies don't have. If your supply is undersized, aging, or lacks the right cable, I'll quote the power supply along with the card so the new card doesn't inherit the old problem.
Should I bring just the graphics card or the whole desktop to Somerville?
Bring the whole tower. A good share of the graphics card complaints I see turn out to be the power supply, the cable, the slot or the cooling around the card. With only the card, I can tell you whether it works on my bench, not why it fails in yours. Leave the monitor and peripherals at home.
Sources and further reading
- Microsoft Learn, Timeout Detection and Recovery (TDR): how Windows resets a GPU that stops responding and the two-second default timeout.
- AMD Support, AMD Cleanup Utility: removing old graphics drivers before a clean install.
- NVIDIA, Ultimate Guide to GeForce RTX 5060 and 5060 Ti: power supply wattage and connector requirements for a current card, including the 12V-2x6 adapter note.
- Tom's Hardware, How to Test Your Graphics Card: benchmark, stress test and monitoring tools.
- Tom's Hardware, How to Check Your Graphics Card Temperature: the sub-90C core guidance and the clock-drop sign of throttling.
- Tom's Hardware, Why Are Power Connectors Melting?: causes of 12VHPWR and 12V-2x6 damage and the case for fully seating the plug.
Graphics Card Repair and Replacement for New Jersey Gamers
Bring the whole tower to 75 N Bridge St in Somerville, no appointment needed, and I'll run the card, the supply and the cooling through a real diagnostic ($75 for a Dell, HP, Lenovo or other major-brand desktop, $149 for a custom-built or boutique PC), credited toward the fix. Then you get a flat-rate quote for the repair, the replacement card, or the supply that should go with it, through my custom PC repair and upgrade service.
📞 (908) 428-9558