Before You Send Your GPU for Repair: Removing the Card, Backups, Photo Records & What to Bring
Key Takeaways
Before sending a card for repair, remove it safely, record its condition, and bring the card only. The card holds no user files, so the real preparation is a symptoms log, fault photos, model and serial details, and the booking number.
First, One Misconception: The Card Holds No User Files
No photos, documents, game saves, or work files live on the card. What the card does hold is firmware, called VBIOS, on its own chip: boot and initialisation code, not user data. What you install, drivers, settings, and game profiles, lives on the PC's storage drive, not on the card. A driver thread on a hardware forum puts it plainly: the driver you install lives on the disk, not on the GPU, while the GPU has code stored in its onboard chips.
The shop's Terms contain a data disclaimer: the shop accepts no liability for any data stored in the card, including game saves, mining earnings, or AI model weights, and advises customers to back up important data before sending a card for repair. Read practically, the position is this: anything of value lives on the PC, and because the shop takes only the card, with the PC staying at home, your files never leave home. The backup that matters for a GPU repair is the record described below, not copying files off the card, since there are none to copy.
The shop takes in cards by two methods only, per its Terms: drop-off at the Shek Mun shop, or SF Express mail-in. All repairs are done at the Shek Mun workshop, not outsourced to third parties. The shop is at Room 1209, Dali Plaza, Shek Mun, Sha Tin, New Territories, phone 2186 8010, open Monday to Sunday, 11:00-20:00, including public holidays. This article covers what to do before drop-off or SF mail-in.
Removing the Card: Seven Steps
Removing a card is straightforward if each step is done in order and nothing is forced. Work on a clear surface, with good light, and give yourself unhurried time.
Step 1: Shut the PC down properly. Do not work on a PC that is only asleep or hibernating.
Step 2: Unplug the power cord. Manufacturer manuals are explicit on this: unplug the power cord from the wall socket before touching any component, and switch off the power supply and detach its power cord before installing or removing any component.
Step 3: Discharge residual power. After unplugging, press and hold the case power button briefly to drain stored charge, and wait a short while so power-supply capacitors can discharge. Community practice agrees this is harmless and commonly done. One limit matters: this does not make everything inside a power supply safe. Never open a power supply. There are no user-serviceable parts inside, and its capacitors can hold charge even when unplugged. Discharge means unplug, hold the button, and wait briefly. Nothing more.
Step 4: Disconnect the cables from the card first. Remove the monitor cables, then the power-supply connectors. For power connectors, press and hold the securing clip, then pull the connector off. Do not yank. A gentle side-to-side wiggle helps if a connector is stiff. Pulling without pressing the clip damages the connector or the cable.
Step 5: Remove the bracket screws, and keep them. Cards are typically secured to the case by one or two screws on the rear bracket. Undo them slowly while supporting the weight of the card, and set the screws aside somewhere they cannot roll away. As a suggestion, put them in a small bag or envelope, separate from the card, so they cannot scratch the board in transit.
Step 6: Release the PCIe retention latch. At the end of the PCIe slot there is a small plastic retention clip. Push it down gently to release. Excessive force can damage the clip or the motherboard. The latch type varies by motherboard, so see the latch section below before pulling.
Step 7: Pull straight out, evenly, without rocking or twisting. Hold the card by its edges, support both sides, and lift it straight up and out of the slot, supporting the card's weight as you lift. If you meet resistance, stop and re-check that the screws and the latch are fully disengaged. Do not force the card. Forcing it can cause permanent damage to the card, the motherboard, or both. Do not use metal tools to pry a stuck card, since metal can short components. If nudging is needed, plastic or wood only.
Support the card's weight throughout: while undoing screws and while lifting it out, never let it hang from the slot by one end.
| Step | Action | Common error |
|---|---|---|
| 1. Shut down | Full shutdown | Working on a PC that is only asleep or hibernating |
| 2. Unplug | Unplug the power cord | Working with the PC still plugged in |
| 3. Discharge | Hold the power button briefly, wait a short while | Opening the power supply, which must never be opened |
| 4. Disconnect cables | Press the clip, then pull; gentle wiggle if stiff | Yanking a connector without pressing the clip |
| 5. Bracket screws | Undo one or two screws while supporting the card; keep screws in a small bag | Letting screws roll away or sit loose against the board |
| 6. Release latch | Identify the latch type first, then release gently | Forcing the latch and breaking the clip or slot |
| 7. Lift out | Pull straight out, evenly, both sides supported, supporting the card's weight | Twisting, rocking, or forcing a stuck card; letting a heavy card hang from the slot |
Latch Types: Check Your Motherboard Manual First
Not every slot releases the same way, and guessing is how clips break. Identify your slot's release mechanism in your motherboard manual before pulling.
The standard latch is a small plastic clip at the end of the PCIe slot, pushed down to release. Some slots use a push-button mechanism instead, and some latches pull toward the rear of the chassis. ASUS Q-Release uses a button near the memory slots that unlocks the first PCIe slot's security latch with one tap, though the button can be hard to press when heatsinks, thick backplates, or other parts obstruct it.
ASUS Q-Release Slim is buttonless: there is no latch to press. The ASUS FAQ describes holding the end of the card nearest the rear I/O and pulling slightly away from the motherboard at an angle to release it. There have been user reports of scratches on GPU PCIe edge contacts after repeated insertion and removal cycles with this design, and ASUS later revised the design. Treat those reports as reports, not as an established fact that the design damages cards. If your slot has a buttonless release, follow that manual's exact motion. If unsure, stop and check the manual rather than improvising force.
The general rule holds for every type: if the card will not move, re-check screws and latch. Do not force it.
Static-Safe Handling
Electrostatic discharge can damage electronic components when they are improperly handled, and can result in total or intermittent failures. AMD's precautions are a good baseline: use an ESD wrist or ankle strap with skin contact, connected to an unpainted metal surface on the chassis; avoid touching the card against clothing; handle the card by its bracket or edges only; avoid touching the printed circuit board or the connectors; put the card down only on an antistatic surface such as the bag supplied with it; and if returning the card, place it back in its antistatic bag immediately. ASUS guidance agrees: use a grounded wrist strap or touch a safely grounded metal object before handling, hold components by the edges to avoid touching the chips, and place removed components on a grounded antistatic pad or in the bag that came with the component. Oracle's guidance adds: do not touch the components along their connector edges.
If you have no wrist strap at home, the minimum fallback is to touch an unpainted metal part of the case first, avoid clothing contact, hold the card by its edges or bracket only, never touch the gold fingers or components, and rest the card only on its ESD bag or an antistatic surface. Be clear about the limit: touching the chassis alone does not ensure adequate ESD protection, and it is not equivalent to a strap. It is a minimum fallback, not a substitute. For transport, place the card in anti-static packaging before it leaves your hands. The shop's own rule aligns: gold fingers go in a protective cap or clean ESD bag, and a bare board must never be wrapped directly in bubble wrap. The ESD bag goes first.
Four Things Not to Do Before Repair
1. Do not disassemble the heatsink or cooler
The shop's Terms allow refusal of devices previously disassembled by unauthorised persons, and unauthorised third-party modification or disassembly is excluded from the shop's warranty. Sapphire states that dismantling the card or removing its heatsink invalidates the limited warranty. The shop's own checklist says the same in practical terms: remove non-original brackets, but do not strip the card down. The cooler stays on. If a custom water block is fitted, bring the original air cooler as well if possible, for bare-board testing. Removing the cooler is the repairer's job. A customer who strips the card risks losing the maker's warranty and the shop's post-repair warranty, and risks physical damage.
2. Do not remove or tamper with warranty stickers or tamper seals
The shop's FAQ states that removing the tamper-evident sticker may affect the maker's warranty, and advises going to the maker first while still in the maker's warranty period. Maker examples of refusal include a burnt power connector, water-cooling leak oxidation, and scratches from repasting.
Brand policies differ, so present this as mixed, not as one rule. Sapphire treats dismantling or removing the heatsink as invalidating the warranty. An MSI forum thread suggests a sticker alone may not void warranty unless damage is caused during removal, and advises keeping the original cooler and reinstalling it for a return. United States law on warranty-void stickers, under the Magnuson-Moss Warranty Act, is United States law only. Do not apply it to Hong Kong. No Hong Kong law claim is made here. The safe course is simple: until the warranty status is confirmed, do not touch the stickers or screws. If the card is still in the maker's warranty period, go to the maker first.
3. Do not flash non-stock firmware or VBIOS
The shop's warranty excludes a customer re-disassembly or flashing of an unsuitable overvolted VBIOS. The reason is practical: VBIOS lives on the card's own firmware chip, and flashing an unsupported firmware can leave the card non-functional. One documented case describes a card that stopped working after a VBIOS flash because an unsupported firmware had been flashed to it. A bad flash can turn a repairable fault into a firmware-repair job, or worse. Leave firmware alone. Record the current driver version instead, as described below.
4. Do not repeatedly power-cycle a burnt or visibly damaged card
If there is a burnt smell, a spark, scorch marks, or a melted connector, stop and unplug. Do not keep switching the system on to see if it still works. A damaged device should not be powered on again until inspected, because powering a contaminated or damaged device can cause shorts, sparks, or secondary damage, and soot and contaminants are acidic and conductive. The shop's own flow page states the same principle: do not blindly power on, since it can cause secondary breakdown of the main chip. Each power-on can feed current through a short and widen the damage. Photograph the damage and send the card in. This is about not repeatedly powering a damaged card; it is not a claim that every single power-on destroys the card.
The Real Backup: What to Record
Since the card holds no user files, the real backup is a record. Five items cover it.
First, a symptoms log. Note when the fault happens, what it looks and sounds like, the exact error messages, the driver version, and recent changes: a new driver, a new game, a new power supply or cable, or a moved PC. Faults are often intermittent. A log lets the engineer reproduce the fault instead of guessing. A video can capture sounds, lights, and intermittent faults better than text alone.
Second, photos or video of the fault. Capture artifacts on screen, fan behaviour such as rattling, not spinning, or full-speed running, black-screen moments, and a Device Manager Code 43 screenshot. The shop's checklist explicitly asks for this, since fault footage helps engineers reproduce the problem.
Third, card identity. Record the exact model, brand plus model, and the serial number from the card's label. Photograph the label rather than only copying the number. Keep proof of purchase or an invoice if a warranty claim may matter.
Fourth, driver version plus recent changes. Note the installed driver version and anything that changed just before the fault: a driver update, a Windows update, a new cable, a moved case. This is PC-side information. The card does not record it.
Fifth, the booking or RMA number. Keep the confirmation screenshot. Without it, the shop cannot match the parcel or the drop-off to your booking.
One more item belongs after the repair, not before it: after the card is repaired, before reinstalling it, use DDU in Windows Safe Mode to cleanly remove old driver residue, then reinstall the latest official driver. That is the shop's checklist advice, and it keeps old driver traces from confusing the fresh repair.
Photo Records: What to Photograph and How
Photograph the card before it leaves your hands. A good set covers the front of the card, with fans and cooler and overall condition; the back of the card, with backplate or board, existing scratches, and stickers or seals intact; a close-up of the serial label, readable in the photo, plus the model label; existing damage, including scratches, a bent bracket, a cracked shroud, scorch marks, or a melted connector; connector condition, including the PCIe gold fingers, the power connectors, and the display outputs; accessories included, photographed together with the card; the fault itself, as photo or video; and handover proof: for mail-in, a photo of the packed box plus the waybill number; for drop-off, keep the RMA or receipt. The shop says it films its own unboxing on arrival, so your pre-send photos are the matching before half of the record.
Why this matters: photos prove what was already damaged before the repair, and a written and visual record avoids disputes over pre-existing condition. Practical tips: use consistent lighting, natural light near a window, and avoid yellow overhead light that distorts colour; use the same angles each time, front, back, sides, and ports; clean the phone lens; tilt slightly to avoid glare; use a plain background. Do not clean damage off the card before photographing it. Cleaning can remove evidence of the fault. Keep originals with their dates. Do not rely on a compressed chat thumbnail alone.
What to Bring, What to Leave Home
| Item | Bring or leave | Notes |
|---|---|---|
| The card itself, in an ESD bag | Bring | The shop takes the card unit only |
| Bracket screws, in a separate small bag or envelope | Bring | Suggestion, not a sourced fact: keeps screws from scratching the board |
| Original box, if available | Bring | The shop's guide calls the original box the best inner packaging if kept |
| Power adapter or cable | Only if relevant | Third-party 16-pin adapter for RTX 4090 or 4080, so engineers can test the cable; standard 8-pin cards need no PSU, since the shop has full test PSUs |
| Proof of purchase or invoice | Bring if relevant | If the card may still be in the maker's warranty but the maker refused |
| Booking or RMA number | Bring | Keep the confirmation screenshot on your phone |
| Fault photos or video | Bring | On your phone; helps engineers reproduce the fault |
| Whole PC, case, SSD, or hard drive | Leave home | Engineers test on the shop's own clean test PC; your files never leave home |
| PSU, motherboard, case, or other parts | Leave home | The shop's guide lists these as do-not-send items |
| Loose screws mixed with the card | Leave out | Keep screws in a separate bag, not loose against the board |
| Any accessory beyond the checklist | Confirm with the shop first | Ask before sending, for example a riser cable, water-block fittings, or extra brackets |
Common Mistakes
| Mistake | Why it matters |
|---|---|
| Forcing the PCIe latch or forcing the card out | Can break the slot, the clip, or the card and motherboard |
| Yanking power cables without pressing the clip | Damages the connector or cable; stiff connectors need the clip plus a gentle wiggle |
| Touching gold fingers, board components, or chips | ESD and contamination damage; ESD can cause total or intermittent failure |
| Resting the card on a bare table, carpet, or ordinary plastic bag | ESD risk; ordinary plastic generates static; rest the card only on an antistatic surface or its ESD bag |
| Leaving the card loose in a bag or box with gaps | Physical and ESD damage in transit; the shop's guide bans gaps |
| Cleaning the card with liquids before repair | Keep liquids away from the card; do not clean it before repair, since cleaning can remove fault evidence |
| Disassembling the cooler or removing stickers | Risks the maker's warranty, the shop's warranty, and refusal of service |
| Flashing non-stock VBIOS | Can leave the card non-functional; excluded from the shop's warranty |
| Repeatedly power-cycling a burnt or damaged card to test it | Can feed current through a short and widen damage; risk of shorts and sparks |
| Forgetting to record the fault pattern | Intermittent faults may not reproduce on the bench; the engineer is left guessing |
| Bringing the whole PC | Unnecessary transport risk, and files leave home for no reason |
| Opening the power supply | Stored charge inside; no user-serviceable parts |
FAQ
Do I need to back up my files before sending the card?
The card itself holds no user files, so there are no files to copy off it. Your files live on the PC's drive, and the PC stays home. What you should back up is the record: symptoms log, fault photos or video, model and serial details, driver version, recent changes, and the booking number.
Can I open the card to look inside before sending it?
No. Do not disassemble the heatsink or cooler, and do not touch the stickers or screws. Opening the card risks the maker's warranty, the shop's warranty, and refusal of service. Leave the cooler on and let the repairer open the card.
How is pickup arranged?
The booking form also lists a doorstep pickup option, but no price, coverage, or timing is published for it, so confirm with the shop first before relying on it.
What if I forgot to photograph the card?
If the card is still in your hands, re-photograph it before handover, following the photo checklist above. If it has already been sent, send it anyway, and describe the condition in the booking form as best you can, and ask the shop about its unboxing record. The shop says it films its own unboxing on arrival, which creates an arrival record. Photos are records, not assurances, and a missing photo does not block the repair.
Do I need to bring my power supply?
No, for standard 8-pin cards. The shop has full test power supplies. Bring a third-party 16-pin adapter only if you use one with an RTX 4090 or 4080, so engineers can test the cable.
What if the card will not come out of the slot?
Stop. Re-check that the bracket screws are out and the latch is fully released, and check your motherboard manual for the latch type. Do not force the card. If unsure, stop and check the manual rather than improvising force.
What about drivers after the repair?
Before reinstalling the repaired card, use DDU in Windows Safe Mode to cleanly remove old driver residue, then reinstall the latest official driver.
What will the repair cost, and how long is the cover?
Inspection is free, and if you decide not to proceed after inspection, no fee is charged. The shop publishes a three-tier reference range of HK$400-HK$2,800, and the actual fee is per the post-inspection quote. Never map a symptom to a tier yourself. The shop's Terms state 90 days of cover for non-man-made damage from the pickup or sign-off date, limited to the specific fault repaired, with the actual period per the quote or warranty card.
Final Check Before You Hand Over the Card
Run through this list before the card leaves your hands: PC shut down and unplugged, card removed without force, card in an ESD bag, symptoms log written, fault photos or video captured, model and serial recorded, driver version and recent changes noted, booking number kept, non-original brackets removed, cooler left on, stickers untouched, firmware untouched, card only in the parcel, no whole PC, no PSU, no loose screws against the board. If any item is missing, fix it before handover. Preparation is the part of the repair you control completely, and it is worth doing properly.