Topic: HELP: Replacing SMD Capacitors on a Radeon 7970 (Read 9044 times)

AndrewK I've got an XFX 6870 with a bad smd capacitor dealing with the power phase controller. Board location for the capacitor is called ac20. Could you point me in the right direction of specs or part numbers, I cant retrieve anything off the previous capacitor as it is ceramic and blown.

At least you knew it was broken before buying it, I bought mine as a good one.

Wish you good luck man 

^^ Hey man, sorry I understood what you were saying... My reply didn't address it though... I will try with a 1x extender... I havn't had much luck with these yet but I'll give it a shot (I think I got two faulty ones... they deffinitely don't work on my 1x slots on board with 2 cards in the 16x).


^I am working on RMAing it but sapphire is pretty specific about warranty being non transferable  (I don't have a receipt or invoice and I'm not sure I feel comfortable completely making up a story)... Plus the card has been taken apart and thermal paste has been reapplied several times.... Also one of the thermal pads for the ram chips was missing so I used a razor to cut extra from around the other pads and applied it to the missing pad.

Thanks guys. If I hear back from athlonmicro and they say I can't RMA I will be desoldering one of the functional looking caps around this area... Checking capacitance and then replacing it with a functional cap of equivalent capacitance from another board.

Sapphire rmas any card under warranty. Just pay the 20 dollar fee and send it to them. Make sure the card looks like it was not totally messed with.

I asked you to try mining with it (in a PCI-E x1) because I have a similar situation.
The card is a 5870 and has a fine crack at the same distance from the PCI connectors as yours but more in the middle.
The crack disrupts some of the circuits and a couple of components are simply missing (they fell off).

Used alone the card is dead as a brick, but when placed in the PCI-E x1 slot using an extender, cgminer is able to see it and mine with it along with the other cards.
It runs at a slower speed (422Mh/s instead of 431Mh/s like the others) but this could be because of the PCI-E x1 slot.

If placed in a normal PCI x16 slot the card is again dead.
Cannot say if in Windows you can do this, I am running BAMT.

 GPU  0:  66.5C 4013RPM | 431.1/430.2Mh/s | A:237 R:0 HW:0 U: 5.97/m I: 8
 GPU  1:  66.5C 3578RPM | 432.7/429.2Mh/s | A:245 R:0 HW:0 U: 6.17/m I: 8
 GPU  2:  66.5C 3716RPM | 422.5/422.8Mh/s | A:201 R:0 HW:0 U: 5.06/m I: 8
 GPU  3:  66.5C 3145RPM | 430.1/429.7Mh/s | A:238 R:0 HW:0 U: 5.99/m I: 8
 GPU  4:  66.5C 4008RPM | 370.7/372.3Mh/s | A:203 R:0 HW:0 U: 5.11/m I: 9
 GPU  5:  64.5C   0%       | 370.5/371.8Mh/s | A:192 R:0 HW:0 U: 4.84/m I: 9

Here is a pic of the actual card with burnt caps.... I will be adding a better picture tomorrow evening.

The small caps belong to pcie lanes, they should be 78 nf if i recall it right. The big caps should be part  of power circuit. Anything over 22 uf should work there. 100 uf should be the best.

However, chip may have got some cold joint if it ran to hot for a long time. If this is the case you nedd to resolder it (hard job).

RMA RMA RMA

I just have pics off the web... But I have seen a bunch of different cards from different generations and this is definitely out of the ordinary. I think once I get a picture up, some of the people with EE/ECE backgrounds can make a more educated guess.

Do you have a working 7970 you can compare the two?

sometimes caps might look burnt but it's from their soldering process....


One of my GPU's looks like it has Flux paste all over a portion of it.....

Well, the "burnt" caps are on the wrong side of the pci-e slot to be a power rail. That said... what do I know? Without a schematic I'm just guessing. The reason I say they are "burnt" is that from visual inspection, they are the only part of the card that does not look brand new. The caps in this area have an oxidized darker look to them. On some of the terminals they are black instead of silver/grey.

The card is at my office right now but when I get it I will take a close up picture of the card so I can show you guys the actual damage.


Thanks for the thoughts and the replies. I think at the very least I will be replacing the caps in that area that I can and see if that helps.

Someone asked about mining with the card. When it is in the second 16x slot windows and RBE identify it. GPUz identifies it incorrectly, but it will not mine and I cannot install drivers on it.

You could always try selling it.  Shoot, I wouldn't mind tinkering with it.   

Just throwing a thought here...have you tried to mine with it (not as the first card) eventually through a PCI-E x1 slot?

I would say it is unlikely you can repair this without access to the circuit diagram or parts.

I can offer some suggestions based on probability.

The caps you have selected do not burn up. When they blow they go open circuit and you cannot tell by looking at them. Seeing similar caps scattered about the board I would say they are filter caps for the power rail. If this is the case it means that the value is not critical so 100nf should be ok. Also if they are filter caps then a power rail is very near.

My best guess based on the limited information is that a semiconductor has blown and shorted out the power rail and I think the power rail heated up and cooked where it was the thinnest and that track is inside the multi layer board so you cannot see it.

If this is the case you have two problems. 1) repairing the track, if it is open circuit you cannot trace it out so you will need a good board to test or a circuit diagram, once you know the route you can link with a wire. 2) you need to locate and replace the component that blew the track in the first place and that could be anywhere on the board....... this is of course if it is a blown track in the first place

All I can tell you for sure is it is impossible for those caps to burn a board. You would have to go high voltage (for their rating)  and the worst I have seen them do is crack with a little pop and if that much of a high volt event happened the rest of the board would be a smoking ruin.

I understand you weren't speaking of schematics... but rather tracing the circuit manually... If anyone happens to know where to get a schematic however I would be much appreciative.


to your point I will try and look up these chips and their values. I have not noticed chips around this area with writing... The board has the capacitors numbered(atleast for the larger capacitors but I assume they are just numbered consecutively as they appear on the board and that this numbering scheme would tell me nothing about their capacitance. Searching these numbers yields nothing in relation to radeon gpus).

I had also thought about the location and whether they were associated with the pci-e interface so I'm glad you mentioned this. I am not an EE so I don't have alot of background with ICs... I have a buddy that is EE PHd candidate... Might have to contact him.

I will post what I can find about the chips and see if anyone has any ideas.

I was not talking about the reference design of the whole card. Look up the small chips next to the burnt out capacitors (if you can read the small print on them). In their data sheets there will be application schematics which you use to geuss more qualified about the values of the burnt out capacitors. Further they are next to the edge connector, meaning some of the are either decoupling for data lines, meaning they can be omittet or they can be subsistuted with any suitable nF value common to pci-e data lines. Or they are power line decoupling capacitors which also can be geussed from pci-e specs.

Thanks for the advice^

Now, I am not a novice to google... But I have been unable to find a schematic for reference 7970s listing current/voltages for circuits or the capacitor's part numbers.

If someone can find this and link I will be cooking with gas.... Also wouldnt mind having these for 5870s, 6970s... etc.

Hi Lookup the nearby ic's, current regulators etc. look through their reference design PDF's and geusstimate the values of the capacitors on your board. trace ground plane and VCC on IC's and passive components with a multimeter. You can be pretty sure that support ic's are implemented as per the chip producers reference design. As said earlier, spending hours reverse engineering a circuit board and spending days sourcing components might not locate the original problem, however doing it the tedious way, will also give you a hint whether the nearby support ic or a current regulator is faulty.

That's my problem... identifying which cap to get... There are so many..

Ok, fair enough.... I have inspected the other side and see no visual evidence of damage.

Do you know how to answer my questions about these surface mount capacitors and matching them, whether I can use previous generations... will I have to check the capacitance... etc?

Thanks,