I Bought 7 BROKEN Graphics Cards to Try and Fix - I Was Disappointed

JLCPCB offers the cheapest PCB with SMT and 3D printing and their circuit boards are very high quality with fast shipping.
Why not start your PCB and SMT experience at JLCPCB from $2.

I Bought 7 BROKEN Graphics Cards to Try and Fix - I Was Disappointed - I paid $3,150 for these 7 salvage graphics cards. Since they're salvage there's a possibility some of them work fine already, some of them are fixable, and some of them are beyond my ability to repair them. I'm hoping to get a good mix but with salvage I just never know.

-----------------------------------------------------------------------------------------------------------

-----------------------------------------------------------------------------------------------------------

Tools I Use:

Equipment Used:

TronicsFix assumes no liability for property damage or injury incurred as a result of any of the information contained in this video. All material provided within this Youtube channel is for informational, educational, & entertainment purposes only. Some of these projects, materials, and techniques may not be appropriate for all ages or skill levels. No information contained in this video shall create any express or implied warranty or guarantee of any particular result.
TronicsFix does not make any claims of the safety of the projects, techniques, or resources listed on this channel and will not take responsibility of what you do with the information provided by this channel. Any injury, damage or loss that may result from improper use of these tools, equipment, or the information contained in this video is the sole responsibility of the user and not TronicsFix. Please do not attempt anything contained in this video unless you accept personal responsibility for the results, whether they are good or bad.

Disclosure of Material Connection: Some of the links in the post above are “affiliate links.” This means if you click on the link and purchase the item, I will receive an affiliate commission. Regardless, I only recommend products or services I use personally and believe will add value to my readers. I am disclosing this in accordance with the Federal Trade Commission’s 16 CFR, Part 255: “Guides Concerning the Use of Endorsements and Testimonials in Advertising.”

The music used in this video does not require attribution.

[Tronicsfix]

As I said at the beginning of this video, I'm no graphics card repair guru. I have no experience trying to fix them so this was just an experiment. If you're looking for good advice on how to repair graphics cards be sure to read the comments on this video as there is great advice from smart people.

[proesterchen]

2080 Tis come with only 11 of the 12 memory channels populated from the factory, also, missing phases in the various VRMs is a normal strategy for the OEM to save money while still using the reference board design.

[ShamblerDK]

1. Never, ever allow GPUs to boot past BIOS without a cooler mounted.
2. That 3060 with the green artifacts is usually an indicator that a RAM-chip is either bad or has lost connection with the PCB. You could try a reflow of the RAM chips.

[Asyq]

1) Don't ever just install and attempt to run a GPU that you know has an issue or is likely to have one.
2) Check for shorts on the power input rails aswell as on the voltage rails the card itself creates (5V, 1.8V, PEX, vCore, vMEM, VDDCI (on AMD GPUs), MSVDD (on all GA102 GPUs except for the 3090Ti)). Each power input is it's own 12V rail, so just checking one will not suffice. vCore, vMEM, VDDCI, MSVDD and PEX have rather low resistances, so check them using resistance mode. DO NOT use continuity mode, or you might just lead yourself down the wrong path. (Getting a close to 0 Ohm reading on vCore on newer GPUs is expected since the GPU silicon has a very low internal resistance.)
3) If you have a short, don't ever just inject voltage, especially if a power input is shorted. First, check if any power can make it's way from that power input directly to the GPU or memory using your multimeter. If there is a direct connection and you just go to inject voltage, you might just be sending whatever voltage you're injecting directly into your GPU or memory, potentially frying them.
4) If you have a short, inspect the componets that are on that shorted voltage rail closely. Don't waste your time looking at componets that aren't on that rail (they can't cause the short). Look for things such as blown capacitors, mosfets, powerstages or a bubbled-up top PCB layer. Look at the solder on mosfets and powerstages, see if there's some small solder blobs or balls around or if the joints look bad and not properly flowed. Those are indications that something got quite warm.
If you have a short on PEX, your core is most likely dead. AMD Vega cards a notorious for having defective cores.
5) If you don't have a short anywhere, run the card. If you have artifacting, it's very likely one or more defective memory banks, mayhaps even across multiple memory channels. For nVIDIA GPUs from Fermi to Turing, use nVIDIA MODS MATS. Run it, and it'll tell you in a plain textfile which memory bank is spitting out errors. You can probably do the same on AMD GPUs and nVIDIA Ampere cards, but I haven't looked into that yet. (As far as I'm aware, the latest version of MODS MATS that is available only supports nVIDIA GPUs from Kepler to Turing). Artifacting can also be caused by the core/memory controller. If the memory controller is dead, you'll likely find discolored underfill around the core. If you see that, the core is dead, or pretty much dead.
6) If you simply have no display, it's very likely just a missing voltage rail, probably caused by a defective logic gate somewhere on the board and/or physical damage (missing capacitor or resistor, GPUs w/o backplates are obviously quite prone to physical damage, so it might just be worth to take a closer look at the back of the PCB). That said, defective memory can also cause no display. In that instance, do step 5.

There are many more things that can cause a GPU to fail POST or not to work correctly, but a short, physical damage or a missing voltage rail are the ones that I myself have seen and fixed the most (by far).

Now, some things in general:

1) Any and all of the discolored PCB and componentry you saw in this video is not caused by excessive heat or overclocking. That's just silicone oil from the thermal interface materials reflecting the light in a funny way. Solicone oil is not conductive or capacitive. It does not damage your graphics card in any way, shape or form.
2) I'm not aware of any GPU that doesn't have some empty pads (and, by extension, missing comnponets) by default. So if you see an empty pad, that doesn't mean someone has used a GPU as a donor for parts.
3) Some GPU models have intentionally missing memory modules to get to the desired VRAM capacity.
4) Pointing a thermal camera at blank silicon doesn't do you anything since the silicon will reflect some IR that's already in your environment and, by extension, throw off your reading.
5) Please don't buy dead GPUs off of random people on ebay. It's a bad idea 'cause most of them are way beyond repair. Only buy dead GPUs on ebay if you can get at least a rough idea of what the fault is and, most importantly, if it's not been messed with already.

[phyro]

I have been fixing gpus for a living for more than a year. Nothing tells me these cards were unfixable.
-The 1080 FTW just had a dead filtering capacitor. This model doesn't have fuses but its probably something minor. There are more voltages than 1V8, 5V and 3V3, there's also FBVDD, PEX and NVVDD.
-The 3060(?) you ran at 140C, if it wasn't toast it probably is since you ran it at 140C...
-No missing ram chip on the 2080ti. Each chip is 1GB, card is 11GB. Missing chip is because they bin the chips that have defective memory controllers and disable that one to have a better yield.

[Irongalaxies]

That 2080 Ti's board isn't missing parts, that's just cost savings done by the 3rd party board partners. Perfectly normal to be missing the last memory module. Check out some of the tear-downs of that GPU.

[Scitch]

6:30 The fans not spinning aren't really an indication. Those newer mid-range to high-end GPUs have what's called a "Zero-Fan mode" wherein the fans will spin when it's first turned on but then go off again until the GPU reaches a certain temperature or is hit with a higher load.

On a usual startup the fans only spinning at the start is not an issue.

[DjPsYcOtIc]

18:49 the missing ram chip is normal, they come like that, even the 1080 Ti's are the same 3rd chip missing. Also to note, when you get no picture and the machine appears to boot, this is an indication that the GPU is fine, but 1 ram chip is needing replacing, there is some software you can get to test your GPU memory and it will tell you witch memory chip needs replacing. Of course this runs in dos from a flash drive etc.

Also if your system boots but has no screen, try using a 2nd card and it should boot into windows on the 2nd GPU and you can see in device manager that the broken GPU 1 will show up as code 21 error, this means the card is working but the memory on the card, one of them chips has died and will give a page in non page area fault, basically use the NVIDIA software tool and find out what chip it reports dead, swap it out, card fixed.

Also if you boot using a 2nd GPU and install team viewer or something like that, log me in, etc etc you can shut it down, take out the 2nd GPU and then boot it up with the broken card, then jump on another PC and login remotely and you will get a very small screen but you can navigate around the system and see that in device manager it has error code 21 device cannot start.

Hope with helps,

Oh p.s. all the other noted missing parts, are normal, its a re-used board from reference to reference I believe.

Good luck and thank you for sharing.

[GeekyGamer]

Well it's clear that this is new territory for you, with the comments on the lack of components on the 2080 Ti (which is definitely a normal thing for GPUs as they re-use board designs for a number of different chips), and running a card without a heatsink on it at all. That made me cringe because seeing 130°+ on the package means the junction temp was probably a lot hotter... For testing a GPU's board for temps a good way to get a view is to either rotate the heatsink 90° and mount it that way if possible, or use a CPU cooler to give yourself a visual of the rest of the board while it's powered on. This isn't a good thermal solution for the power management and memory though, so I wouldn't recommend doing any stress tests with this type of setup, but it's good for checking the rest of the board when voltage is applied.

With most GPUs now, they turn on the fans for a brief moment after boot to give a kind of "sign of life", then turn them off as the board boots up and starts controlling the fan speed based on temperature. At low enough temps, the fans will be off, hence why you see the fans turn on and then shortly back off when turning the pc on.

[Alvin]

Couple of tips for people new to GPU repair: before plugging the GPU into a working system, make sure none of the major power rails are shorted, or you might end up with more dead hardware than you started with.
Also with the GPU prices recently, it's reasonable to assume that any "expensive" card sold for salvage has already been diagnosed and either deemed a no-fix or made worse (some people get desperate and put them in the oven). Get older NVidia cards like 700/900 series to get started, much higher success chances. It's usually harder to find info for AMD cards, so stick with NVidia until you get a general feel what all the parts on the card do.

[error]

18:34 that's how that card is made, it's not missing any components. Also by looking at what they did with thermal pads there's no way they could remove components so cleanly.

[mpla]

It's normal for graphics cards to get hot in random places especially on memory chips so heatspots don't mean the end of graphics cards. The processor on the 3060 that is discoloured is most likely bad so it is most likely unfixable. Also you should test the 12 volt 5 volt 1.8 vmam vcore for continuenty. If it shorted to ground you can apply voltage and you will see with the thermal cam . If the memory is shorted it's a lost cause. The second doesn't seem too bad but not unfixable since all the voltages are present is most like a 0 ohm resistor. You could have checked with the thermal cam for a heatspot while being connected( if there is no short the thermal cam will show nothing ). In graphics cards While using the thermal cam you should know that the heat is very very minimal you will hardly be able to see it(random spots might get hot but when something is faulty on GPUs the heat is low, unless its the processor.).

[ApocDevTeam]

The problem with dead graphics cards is that when there's a catastrophic power failure in one of the components, there's a high chance it will fry a lot of other components in the process (which may not be visually apparant). A lot of times the GPU core and/or memory modules get destroyed as well. They are extremely sensitive to voltage spikes. Even a couple of volts too high can instantly destroy them. There's also a lot of annoying issues that can happen like faulty memory modules which you have to individually test for and then replace the module. In general graphics card repair is a huge pain and only recommended if you really specialize in it and know all the little details of what's happening with each generational architecture / board design. I know a lot of repair shops avoid graphics cards, also because there's so many different models that it's hard to get replacement parts. According to NorthRidgeFix over 50% of graphics cards they work on end up being a "no-fix" either because the problem cannot be found or the GPU core is dead. Big time waste.

[MMKnight_]

The "missing" components on the 2080 ti often means that it is using the same pcb as a higher tier of gpus/a better version of that card that needs those extra components.

[rambofan]

It doesn’t matter what Steve is fixing I’m gonna watch it and it will make my day better.

[MrCrabs]

I wouldn't trust a card that's only stability test was heaven.

[lagautmd]

What always amazes me about these is learning about the insane number of components besides just the ICs. For these to work properly all these capacitors, resistors, mosfets, etc. must work within tolerances and not fail. Even with a tiny chance of failure per year the probability of a failure somewhere becomes very high. So, the chance of failure for these individual components must be near non-existent. That's amazing manufacturing quality in fundamentally inexpensive components. That means the entire supply chain making these components must be extraordinarily tight on accuracy and lack failure points. It quickly becomes bewildering.

[borisgospodinov]

2080Ti is not missing a chip. That's how high-end GPUs SOC is configured to work/use the amount of VRAM with the bus. Make a research first. Same with the missing VRM parts.

Same apply for 1080 Ti and 3080++

[ZackMuffinMan]

Depending the card, the fans won't spin up until needed. Also, I wouldn't run them without at least a fan on them. They heat up quick (within seconds) and can cause permanent damage.

[jmk]

With the 3060 perhaps consider reducing the Memory clocks with MSI afterburner. On the 2080ti with "missing parts"- they're probably not missing, the board design is probably just reused between different models. You could try a heat gun on the non working cards, but GPU's are difficult to fix even for people who know what they're doing.