I guess it depends on how tenacious you are. Testing 10s to 100s of MHz signals is non-trivial, especially on custom ASICs with (at least nearly) zero public documentation on what the pins should be doing. You can reverse-engineer some of that.. clocks, maybe some of the memory stuff if they're standard footprints and/or you can get the datasheets for those ICs -- but a lot of those are locked under NDA too.

I've always seen the BGA rework stuff as a "might as well try and see what happens" thing. BGA solder joint failure is a common enough failure mode. Maybe that's all it is. But who knows.

So, for me? Not worth the trouble. I pretty much stop at replacing caps. If (e.g.) discrete transistors are failing, I would start questioning why that happened, and if it took anything else with it, and that's a lot more than I want to deal with.

OTOH, I have been curious what it would take to tap into the digital side of the RAMDACs of older cards and replace them with a DVI transmitter.

Hi SirNickity
I wasn't so much thinking of testing digital signals on custom ASICs. From the sort of repair I've done for example on miniature SMD circuitry on diverse devices such as Android and MAG IPTV boxes, or for that matter SATA Hard Drive controller boards, It is generally more a matter of testing the things that can be checked such as all the various power rails, oscillators, reset signals etc. if a Processor has the correct power, has a clock and isn't held in reset and it still isn't working then probably I can't fix it.

But at least half the time I find for example short circuit SMD capacitors, short circuit or otherwise faulty Buck regulators, blown or short circuit power FETS, duff electrolytics, common faulty components such as HDD motor controller IC, duff PWM controllers and duff Gamma correction ICs on LED/LCD TCON boards..... that sort of thing. For those components you can almost always find the datasheet/pinout and can always rework them with hot air, some good solder skills and some decent bench magnification.

From what info I can glean so far regards about video cards is that they are fairly simple devices with a few sections - at lease the more modern ones. Basically they seem to be 50% buck regulator circuits, and the other 50% made up from RAM, the GPU itself, a Bios (ROM), possibly some 'glue logic' to interface with the PCI-e or AGP slot to and some circuitry to monitor errors such as temperature or over current. That seems to be about it.

So now i am wondering how often that circuitry (especially the buck regulators) is failing compared to how often it is the GPU. Power supply isn't so hard to diagnose/repair and even RAM should be at least 'doable'.

Older cards i would think are less integrated, for example having separate RamDAC instead of integrated in the GPU.

Many tutorials seem to say that the only thing that ever goes wrong with a video card is the BGA reflow soldering. So they are easy to repair - throw it in the oven that will fix it. Don't even bother to see if the correct supply voltages are present to the GPU and RAM first, why would you even want to bother to check for something simple like that. Remove the plastic parts and GPU heatsink first? Nahhh....

Oh and put it at 200C for 20 mins. That is hot enough to melt solder....

Well actually it isn't... lead free solder which is what you will find on your GPU doesn't even melt until 220C. Personally I find those tutorials just plain stupid. 😲

So anyone here been repairing video cards and have some idea of the proportion of actual GPU failures to the proportion of other more easily fixed problems?

I have been having component level repair work as a hobby for past 10 years. I mostly do modern apple products but every now and then i work on something more interesting like the Voodoo 5 5500 in my other thread.

dicky96 wrote:

From what info I can glean so far regards about video cards is that they are fairly simple devices with a few sections - at lease the more modern ones. Basically they seem to be 50% buck regulator circuits, and the other 50% made up from RAM, the GPU itself, a Bios (ROM), possibly some 'glue logic' to interface with the PCI-e or AGP slot to and some circuitry to monitor errors such as temperature or over current. That seems to be about it.

This is what I was thinking. If it's not something jellybean like a VR, you're probably about out of luck. I wouldn't expect those to go bad too often, unless the board were subjected to really rough conditions (which would imply other components are probably also suspect.) But hey ... I would love to see someone make it happen if anything CAN be done to salvage failed cards. Obviously can't help much ... but I wish you the best of luck! 🤣

I've recapped my Radeon 9600XT card and added 2 caps to my VooDoo2 card (never done that before a day in my life) back in December (Christmas Eve/Day). As for reflowing the video chip, I've never done that before and a lot of people recommend a heat gun and flux to reflow the solder underneath the GPU chip itself. I've never done traces before on a video card as I never had a trace scuff up on me, and I also clean the contacts off for the video card since sometimes an issue would be dirty contacts that connect to the AGP, ISA, PCI, or PCIe slots. Once a GPU has artifacts, I replace it with a spare or a new one.

bjwil1991 wrote:

As for reflowing the video chip, I've never done that before and a lot of people recommend a heat gun and flux to reflow the solder underneath the GPU chip itself.

OK from my attempts to reflow large BGA on other devices I have come to the decision a cheap hot air station such as the one I have (the 858D which is quite common https://www.ebay.co.uk/itm/220V-LCD-858D-Hot- … 3.c100505.m3226 ) just does not have enough power to get the solder under the GPU to melt.

I believe that using a pre-heater to get the whole PCB up to about 150C by heating from below, and then using a hot air station will do the reflow job but I don't have a preheater at the moment. Hot air preheaters are the best type apparently (compared with infrared) and are only about £40. https://www.ebay.co.uk/itm/New-Kada-853B-Preh … a4RlP:rk:5:pf:0

Also from research and online reviews it appears that a good quality hot air station such as the Quick 861D https://www.ebay.co.uk/itm/QUICK-861DW-Hot-Ai … cPZF2:rk:3:pf:0 is up to the job but that is more expensive than the cheap hot air station plus the preheater.

Anyway as I said earlier I am interested to find out how many faults are Power Supply related and not the GPU itself.

It seems the only way I am gonna find out is buy some 'spares or repair' graphics cards off ebay and use them to learn. I don't mind chancing some money if I learn new skills. It could be well worth the money/time. OK will keep you all informed what I find out. Can't easily get any faulty GPUs here in the Canary Islands but will get some various types when I am back in the UK for a week early February.

Thanks for the Good Luck wishes. Lets see.

Ric.

SirNickity wrote:

This is what I was thinking. If it's not something jellybean like a VR, you're probably about out of luck. I wouldn't expect those to go bad too often, unless the board were subjected to really rough conditions (which would imply other components are probably also suspect.)

SirNickity I just figured out what you meant by 'something Jellybean like a VR' You mean Voltage Regulator?

No a Buck Regulator is not quite the same beast at all. A Voltage Regulator would generally refer to a Series Regulator which would for example take a 12V supply at 1 Amp, generate a stable 5V supply at 1 Amp and dissipate the remaining 7V x 1A = 7 Watts as heat. A Buck Regulator is a very efficient DC to DC converter and can step the voltage down or up.

A Buck regulator on a graphics card for example would take 5V at 20 Amps and generate 1V at 100 Amps for the GPU. Both the input and output are 100 Watts (5V x 20A = 100W, 1V x 100A = 100W) but the voltage and current are different. This of course is a theoretical buck regulator circuit with 100% efficiency - in reality there is some loss of efficiency which is dissipated as heat but real efficiencies are still very high, often over 90%

Voltage regulators are pretty damn reliable I agree. Buck regulators or more correctly Buck Converters are far more prone to failure, or even self destruction, in my experience.

From what I looked at so far most graphics cards have several Buck Converters giving a number of separate voltage rails - one for the RAM power, one or more voltages for the GPU, and another for the Glue Logic. Also the power circuit for the GPU has several phases (which means several converters in parallel but switching in different phase at very high frequency) to provide the huge current at low voltage. This is where I wonder if a lot of faults actually occur.

Rich

Most failures on Graphics cards that I have seen (on ANY brand) is RAM related. Artifacting cards from Nvidia, ATi, 3dfx etc. are very common so if you can replace RAM chips you can save a lot of cards.

Sometimes artifacts can be due to GPU failures/bad connections or bad capacitors but memory chips are tending to fail a lot on graphics cards IMO.

Because of the high age capacitors become more and more a problem but it can be solved fairly easy.

Hi meljor.
That's interesting. Are you finding that kind of RAM failure on older cards or just on faulty cards in general?

best regards

dicky96 wrote:

From what info I can glean so far regards about video cards is that they are fairly simple devices with a few sections - at lease the more modern ones. Basically they seem to be 50% buck regulator circuits, and the other 50% made up from RAM, the GPU itself, a Bios (ROM), possibly some 'glue logic' to interface with the PCI-e or AGP slot to and some circuitry to monitor errors such as temperature or over current. That seems to be about it.

You got that right, basic structure hasn't changed since early AGP/PCI video cards that used separate voltage regulators for RAM and GPU.

dicky96 wrote:

So now i am wondering how often that circuitry (especially the buck regulators) is failing compared to how often it is the GPU. Power supply isn't so hard to diagnose/repair and even RAM should be at least 'doable'.

GPU die doesn't fail on itself, it fails due to external factors: overheating due to over-voltage/insufficient cooling, ESD or over-current. Its power supply circuit (multi-phase synchronous buck converter) is less reliable and tends to fail more often.

dicky96 wrote:

Many tutorials seem to say that the only thing that ever goes wrong with a video card is the BGA reflow soldering. So they are easy to repair - throw it in the oven that will fix it. Don't even bother to see if the correct supply voltages are present to the GPU and RAM first, why would you even want to bother to check for something simple like that. Remove the plastic parts and GPU heatsink first? Nahhh....

I'd say that 80% of all failures are due to cracked, rubbish, lead-fee solder of BGA packages and tell me more about planet conservation with that ROHS compliant, lead-free solder, that in the end, only causes an increase in video card demand and production...

dicky96 wrote:

Oh and put it at 200C for 20 mins. That is hot enough to melt solder....

Not only that would damage plastic parts, but also semiconductors.

dicky96 wrote:

Well actually it isn't... lead free solder which is what you will find on your GPU doesn't even melt until 220C. Personally I find those tutorials just plain stupid. 😲

That depends on used alloy only, manufacturers tend to use alloys, that melt below 200*C.

dicky96 wrote:

So anyone here been repairing video cards and have some idea of the proportion of actual GPU failures to the proportion of other more easily fixed problems?

Few out of interest only, diagnosing faults is easy, once you grasp the concept of how particular device functions, but most of the time, it's either cracked balls or dead voltage regulators.

I currently possess a 8800GTX, that has a dead RAM voltage regulator circuit, which was killed by one, shunting to ground, DDR3 SDRAM 😵
Tested card by powering memory block with an external voltage regulator, which showed that GPU was OK.

dicky96 wrote:

Hi meljor.
That's interesting. Are you finding that kind of RAM failure on older cards or just on faulty cards in general?

best regards

Doesn't matter what type/brand, I have seen a lot of cards that gave weird lines/checkered board patterns/artifacts etc. etc. which is often regarded to as ''ram problems''. Only on a very few occasions down clocking the ram worked for a short while but most of the times it didn't help.

It does seem to be a more common issue from cards like the Geforce 4 Ti series and later cards, and much more problems with the high-end ones ofcourse (like the ti4600). I don't think overclocking was that popular around the time of geforce3 and lower so that might also be a factor? Coolbits/Rivatuner etc. made it very easy for the masses to push the cards a bit further..

I believe that when you can replace ram chips you can revive a lot of cards. But maybe I'm wrong and it is simply bad connections/solder/voltage regulators, who knows?

dicky96 wrote:

SirNickity I just figured out what you meant by 'something Jellybean like a VR' You mean Voltage Regulator?

No a Buck Regulator is not quite the same beast at all. A Voltage Regulator would generally refer to a Series Regulator which would for example take a 12V supply at 1 Amp, generate a stable 5V supply at 1 Amp and dissipate the remaining 7V x 1A = 7 Watts as heat. A Buck Regulator is a very efficient DC to DC converter and can step the voltage down or up.

I meant voltage regulator as "an IC what regulates some input voltage to some output voltage" without any regard as to how it does such a thing. Contrast to a current regulator which modulates voltage into a variable resistance to obtain some target current. The example you gave above is what I know as a "linear voltage regulator." A buck regulator (which can only convert down), and a boost regulator (which converts up), or even a SEPIC (which converts all over), etc., all fall under an umbrella of "switching regulators", which can be rolled up into a larger category of "PMIC" - general power management ICs. 😁

dicky96 wrote:

Voltage regulators are pretty damn reliable I agree. Buck regulators or more correctly Buck Converters are far more prone to failure, or even self destruction, in my experience.

Hm, I have no personal experience with failure statistics, but I could see that. Likely due to marginal design, I would wager. With a linear reg, about the only thing working against you is heat. With switching regs, there are many other variables - for example, the coil and diode and resulting flyback spikes, or if the regulator relies on an external switch, then perhaps gate capacitance or inadequate damping are to blame. In theory, the IC itself ought to be fairly resilient, but there is certainly more room for error. Probably all to do with cost-cutting measures, in the end.

OK SirNickity, you meant Voltage Regulator of any type. Actually I just came across VRMs and did some reading up about those. Nothing I have worked on before used these AFAIK.

I've grabbed some spares or repair video cards off ebay. I now have 12 of them coming for an average price of £5 each inc postage. They are a mix of AGP, PCI and PCI-e. So I'll have a look at them and hope it is a representative sample of older video cards. I may grab one or two more recent faulty GPUs as well if I see them at the right price. The higher end modern ones seem to go anywhere from £40 to over £100 spares or repair, so some folks must be successfully fixing these otherwise where wouldn't be a demand, however looking at sold listings sometimes a recent model goes through cheap for some reason.

I have to wait until first week of Feb when i go back to the UK so I can pick them up.

Rich.

I've bought many untested cards in the past for very cheap in some local markets (less than 5 euros). I've had many cases when they had artifacts. And those artifacts looked like a faulty modern PC Ram kind of issue, hence I think those cards with artifacts have bad memory chips on them.