Graphics Card VRAM Chip Replacement Procedure

Last Updated – 14/08/2022

This is a basic procedure for how to safely remove/replace VRAM chips. This procedure uses a preheater to warm the whole card, but many people also successfully remove/replace VRAM chips using a hot air station alone (which often requires a higher hot air temperature). It always looks easy when watching experts do this procedure, and with experience, probably it can indeed become very routine. However, like most procedures, there are risks and it’s easy to underestimate the value of experience. So, if it’s the first time you’re trying it, I recommend practising until you feel comfortable on less valuable or scrap cards. I am learning too and trying to refine this procedure as I go, so all thoughts are very welcome, please do add comments.

See also Graphics Card VRAM Reball Procedure

TODO

  • Add a section on missing and damaged pads
  • Add note on the case where multiple chips report errors

Prep & Tools

I would say the following tools are essential.

Essential

  • Hot air station – I use the heater without any nozzle fitted. Better quality stations can quickly pay off.
  • Soldering iron with flat tip e.g. bc3 or similar for cleaning pads
  • Flux – I use a syringe dispensed, gel-type flux e.g. AMTECH NC-559-ASM, or Mechanic AD-559
  • Leaded solder & solder wick/braid – For cleaning pads following removal. I typically use a 1mm braid. Wider braids can also work, but be careful not to let it stick and rip pads when cleaning!
  • A suction tool or tweers for chip removal

I also use the following optional tools. It’s possible not to, but I often find them very helpful.

Nice To Have

  • PCB preheater – Nice to have, allows for lower hot air temps and more margin for error. However, it’s completely possible to successfully replace VRAM chips without a preheater (many do).
  • Microscope or some form of magnification – Magnification is often useful for cleaning and inspecting pads following chip removal. A microscope can become essential in the case where pads must be repaired.

Before You Start

  1. ALWAYS Have a measured position become getting to work:
    1. If possible, save any test results from analysis software (e.g. MATS, tserver) before making a start. Even if you already know a certain chip has errors, it can help to know the state before changing anything.
    2. Measure the resistance of the Vmem rail before and after (don’t let shorts make things harder!).
  2. GOOD TO cover nearby sensitive components (e.g. electrolytic capacitors) with foil or Kapton tape.
  3. GOOD TO make a quick plan for chip removal. For example, there are often small capacitors surrounding VRAM chips (consider taking a photo for easier replacement in the event they get dislodged).

General Tips

These tips are of course based on my experience, please don’t take them as expert advice. I am improving, but this is mainly to share the mistakes I have made.

  1. GOOD TO clean pads thoroughly and carefully. Plenty of flux. Sweep the pads with a soldering iron with a blob of leaded solder first, trying not to drag the iron on the actual pads, and let the solder ball glide.
  2. BE CAREFUL not to overheat the chip i.e. if using higher heat (e.g. 400-450 deg C), be quicker, if using moderate heat don’t spend too long. The chip can get damaged. I have baked chips by overdoing the heating i.e. spending too long.
  3. GOOD TO commit to the ‘side nudge test’ to confirm solder is flowing – I personally had a problem doing this (I feared that I would dislodge the chip or cause solder balls to fuse). However, it’s probably worse to try to guess if the solder is flowing and often wastes time by having to repeat the process. Without the ‘nudge test’ it is easy to underheat (see example here MSI GTX 1050 2Gb) or worse, possibly overheat the chip.
  4. TRY NOT TO use loads of flux when soldering the new chip. Just a thin layer all over the pads is enough. Too much flux seems to lead to a higher chance of fused solder balls. I used to do this a lot. It also doesn’t seem necessary to add lots or any flux when removing the chip, doing so isn’t a big deal, but tends to create more mess and smoke.
  5. GOOD TO clean the desoldered pads with IPA while the card is still on the preheater (or while the card is still hot), it makes the flux residue much easier to remove.

Not happy with ‘nudge test’ to check when a chip is soldered? Try the ‘deliberate misalignment’ test!

One problem with the ‘nudge test’ when re-soldering a chip is that it is quite possible to do it too early and mess up a partially soldered chip. Another option, which I am starting to prefer, is to ‘very slightly’ misalign the placement of the replacement chip e.g.

Deliberately misaligned VRAM chip before resolder

Obviously, the misalignment must be such that the solder balls under the chip are still in enough contact with the pads to pull the chip perfectly straight on re-flow e.g.

VRAM chip perfectly aligned after reflow

As you can see, those balls know exactly where they belong when re-flowed and you know the chip is back in place!

Chip Removal

No preheater? – This procedure assumes you have a PCB pre-heater. If you don’t, my approach would be to use the hot air nozzle to slowly manually pre-heat the board and especially the area around the VRAM chip by working the nozzle about from a slight distance e.g. 10cm for say 30 seconds. This should heat reduce thermal stress (e.g. expansion of one specific PCB area) and possibly aid removal. Note, this is just as an idea, the actual timing and heat will vary with PCB size and may not even be necessary for smaller boards.

  • Preheat the board to about 130-150 underneath and 110+ on the surface (can vary a bit) and cover any sensitive components with Kapton tape or kitchen foil.
  • Apply flux around the chip (optional)
  • Begin to slowly heat with hot air station set to 375-425 deg C depending on the side of the PCB, max 450 (very large PCBs, very stubborn chips or perhaps without preheater)
  • Try to raise the temperature over say 30 seconds to 250+, the flux should start to bubble
  • Then close in and keep heating until it’s possible to slightly nudge the chip easily from the side
  • remove chip with the suction tool

Pad Cleaning

  • With the board still on the preheater and a BC3 soldering iron at 300, add more flux and run the soldering iron gently across the pads and wick gently to get the worst of the solder off
  • Add some leaded solder to the iron tip and gently ‘float’ the solder iron tip across the pads. Cleaning the tip off in between sides.
  • Carefully wick away as much solder from the pads as possible (be careful with wider wick not to let it stick and potentially rip off pads).
  • (Optional if you like to clean first) With card still on the preheater, clean area with IPA using a cotton-bud
  • Ideally, you need the pads as clean and free of solder as possible. If they aren’t it may still be fine, but it could possibly increase the risk of excess solder causing ‘solder bridges’ and possibly shorts.

Chip Replacement

  • Preheat board as in chip removal above.
  • Add a thin layer of flux to the pads.
  • Try to align the chip as well as possible.
  • Slowly heat as with chip removal.
  • Set the hot air station to 375-400 deg C (depending on PCB size mainly) and close in until the chip settles down and repeat the same gentle ‘nudge test’ as in removal.
  • (optionally) clean up any flux residue, a lot easier whilst the board is hot.
  • Allow the board to cool before testing.

Testing

  1. (before powering on) When cool, check the resistance of the Vmem rail is normal. If shorted, the replacement procedure can be repeated (see common issues). If too high (like very high for memory), then it could be not enough was applied or uneven heat resulting in the chip not being fully soldered, you can reflow in this case.
  2. Power on test, if any usable picture is shown (may have artefacts), power down and replace the cooler before further testing.
  3. Consider retesting with mats or tserver.
  4. If all is well, proceed to windows and continue to stress-test etc for validating the fix.

Common Issues

These are issues that I have personally experienced.

  1. Short on VMem rail – One or more pairs of solder balls may have likely joined, possibly there was too much old solder left on the pads after cleaning (see HD 7850 – Attempt 1). Possibly too much flux might increase the risk of this. The replacement must be repeated.
  2. MATS errors on every bit – If this wasn’t the case before replacement, the chip may not be soldered down correctly (e.g. insufficient heat/reflow of solder balls). Example GIGABYTE GTX 1050TI 4GB Card B. A successful reflow may fix this.
  3. Artefacts – If there were previously no artefacts, then this can be a sign of one or more loose solder joints. A possible cause might be that not enough heat was used to get all solder balls correctly soldered. A successful reflow may fix this.
  4. Exact same MATS errors after successful replacement – memory controller faults, faulty connections under the GPU core and broken tracks can also cause MATS/memory errors. This needs more discussion and examples, but this may be identifiable from the patterns and exact errors in MATS with experience. Resolving this is likely beyond this procedure.
  5. Damaged or missing pads – After chip removal, sometimes pads will be damaged or missing. This doesn’t always indicate a problem, as certain types of VRAM chip can have pads that are not connected – for example, see XFX RX580 8Gb (Card A). However, essential pads can be damaged before or during chip removal and cleaning, in this case, they must be repaired (TODO add separate guide on VRAM pad repair)

Samples & Results (cases where there were issues)

These were some of my first attempts, when I look back now, my pad cleaning skills were relatively poor and I had issues as a consequence e.g. ripped pads from solder braid sticking, shorts from too much old solder being left on the pads before replacement. Hopefully, these examples can help others from repeating my mistakes.

HD 7850 – Attempt 1 – Failure (short / suspect too much solder left on pads)
HD 7850 – Attempt 2 – Cleaned with leaded solder – May have damaged a pad using too thick wick – Chip replacement was still successful and the card was reliable (looking back, the cleaning could still be better on some of the pads).
MSI GTX 770 Card B – Attempt 1 – Artefacts and mats show the same chip has errors. I can see now that there was too much old solder left on some of the pads, but I think the main issue was more that I didn’t reflow the solder sufficiently (Initially, I had a tendency not to do the ‘hudge test’ for fear of dislodging the chip. It is important to do this to ensure the solder is flowing, especially on these larger PCBs)
Zotac GTX 1060 – Attempt 1 – Initially had a picture, then a blank screen with backlight once cooler was replaced, mats shows errors on the same chip. Whilst this example wasn’t perfect, the actual issue with card Zotac GTX 1060 6gb is more likely the BGA under the core.

Links to example cards (successful fixes)

Here you can take a look at some examples of cards that have been through the above process.

  • Sapphire HD 7850 2Gb (successful replacement, probably luck, as it was my first one)
  • MSI GTX 770 2GB (2048 MB) Twin Frozr (A) (Initial reflow failed, eventual chip replacement for a hopefully long-lasting fix)
  • MSI GTX 770 2GB (2048 MB) Twin Frozr (B) (similar to card A above, the initial reflow failed and the chip needed replacing. These cards helped teach me that reflows should not be considered ‘permanent fixes’)
  • Zotac GTX 1060 6gb (initially a successful VRAM replacement, then the card developed memory errors again after stress. Suspected core BGA issues, I plan to re-ball the core on this card when I get equipment and a chance)
  • MSI GTX 1050 2Gb (various self-inflicted issues, a good one to look at for some examples of mistakes! 🙂 Ended up re-balling the chip for this one – successful)
  • GIGABYTE GTX 1050TI 4GB Card B (successful replacement, one of my best ones so far)
  • MSI GTX 1060 3Gb (successful replacement, apart from some capacitors that I accidentally dislodged!)
  • XFX RX580 8Gb (Card A)
  • TODO add other examples

2 thoughts on “Graphics Card VRAM Chip Replacement Procedure

Leave a Reply

Your email address will not be published. Required fields are marked *