On September 2nd, a customer who did interior design came to the store with a host: the graphics card was a GALAXY RTX 2070 general bought in 2019. Recently, he played Black Myth: Wukong frequently, and the screen flashed back. Snowflake color appeared on the screen. The 3DMark score dropped from 9800 to 5100. He checked it himself and said online that "the memory is not welded, it is better to change" to ask if it is worth saving.
Step 1: Determine the type of garbled screen. Bad core and bad memory are two different things.
Flower screen is not a disease, at least divided into three kinds, the treatment plan is completely different:
| phenomenon | high probability cause | repairability |
|---|---|---|
| Regular color block/checkerboard, specific angle appearance, running score stable decline | Memory particles are not welded or damaged | Repair, welding or particle replacement |
| Full screen snowflakes, artifacts worsen with temperature rise, sometimes can be turned on sometimes cannot | Memory solder (BGA solder fatigue) | Repairable, case in point |
| Driver error Code 43, core area black, power-on no display fan crazy rotation | GPU core damage or power breakdown | It's hardly worth fixing. |
Let's start with a non-destructive scan: Remove the graphics card and use an eraser golden finger, insert the single card directly into the first slot of the motherboard, change the driver to 566.36 and reinstall it cleanly. The phenomenon remains the same-eliminate contact and drive problems. Then use the memory detection of MATS idea (for NVIDIA cards, we use OCCT VRAM test + 3DMark stress test cycle), it was found that in the third round of stress test, snowflakes began to appear when the core temperature reached 71 ° C, and disappeared after cooling down.Temperature-dependent, reproducible, concentrated in specific areas of the screen-typical memory solder.
Step 2: BGA soldering, temperature curve is the core
The memory particles are BGA packages (bottom ball grid array), and the weak solder is a micro-crack between the solder ball and the PCB pad. After thermal expansion, the crack becomes larger and leads to poor contact. The principle of soldering is to use controlled heat to melt the solder again and heal the crack. This step is most afraid of two things: insufficient temperature (white dry) and temperature runaway (blow the core or let the PCB bubble).
Our operation process (amateur players are not recommended to imitate, no infrared temperature measurement is blind blowing):
- Remove cooling and remove residual thermal paste, exposed PCB. High temperature tape around the memory protection, the core wrapped with tin foil reflection heat.
- flux penetration: Spot a small amount of BGA flux (AMTECH NC-559) around the suspicious particles, and use capillary action to penetrate into the solder ball gap. Do not wash the board with water to prevent the flux from being washed away.
- preheat: The air volume of the hot air gun is adjusted to 3, the temperature is 180℃, the whole board is preheated for 90 seconds at a distance of 15cm from the board, and the infrared temperature measurement confirms that the back of the PCB is 80-90℃. The preheating is not in place, and the direct high temperature will make the PCB heated and deformed.
- backflow: Change the air nozzle to a small diameter, the temperature rises to 320℃, the distance from the memory particle is 2cm, the circle is heated for 25-30 seconds, the temperature measuring gun is fixed on the particle surface temperature to 235-245℃ (GDDR6 lead-free solder reflow window), and the natural cooling is 3 minutes. During this period, the board will not be touched.
- Process all 8 memory cards one by one, 6 on the front and 2 on the back. Rather than bet on which one is bad, it is better to remelt them all again, and the work time is only 20 minutes longer.
- rinse: Absolute alcohol + antistatic brush to clean flux residue, dry.
Step 3: Install and accept, burn-in test 4 hours
The core thermal paste is used with Xinyue MX-7921 (originally dried into powder), and the memory thermal pad is replaced by 1.5mm (the original pad has collapsed). Acceptance criteria after installation:
- 3DMark Time Spy runs back to 9700+(5100 before maintenance);
- OCCT VRAM test 1 hour 0 errors;
- FurMark burn-in test 30 minutes, core temperature 74℃ stable, no garbled screen;
- "Black Myth" actual game 2 hours (1440p high definition), no flash back no snow.
Finally, the whole machine is delivered in 4 hours for continuous burn-in test. Total cost: man-hour fee RMB 200 + thermal pad/thermal paste/flux consumables RMB 80 = RMB 280. The customer compared it on the spot: RTX4060 new card with the same performance RMB 2300, refurbished 2070 also RMB 900 and the color is unknown. This RMB 280 is worth it.
Which cards are not worth fixing? Our criteria
- Core bad: Drive Code 43 + core power supply short circuit, repair success rate is less than 20%, and repair is also easy to relapse, direct persuasion.
- Residual value too low: GTX 1050Ti card with residual value less than RMB 200, RMB 280 repair fee is not cost-effective, it is recommended to exchange.
- Heavy usage history of mineral cards: memory long-term 95℃+ operation of the mine card, solder joint fatigue is comprehensive, after welding 3-6 months recurrence rate is high, we repair but will clearly inform the risk.
- PCB blistering or burn marks: Water, burned boards, multilayer internal damage cannot be repaired.
Guangzhou graphics repair market price reference: dust cleaning for thermal paste RMB 80–150, memory plus solder RMB 250–400, single memory replacement (including particles) RMB 350–600, power chip-level repair RMB 300–800. ACCPC repair graphics card are first tested after the quote, no charge for testing, repair burn-in test 24 hours before delivery, plus solder warranty 90 days.