Replacing faulty RAM on video cards?

[Agent24]

I have an old Radeon X300 which has what is considered a classic sign of bad VRAM; lines down the screen.

I'm thinking of replacing its RAM chips, but wonder if this is actually likely to fix it?
Has anyone done so, and if so, did it work?

Could it be simply bad solder joints, or that one of the several ceramic capacitors next to the RAM chips has gone shorted? Or something else?
Does anyone have experience with repairing RAM faults, and knows the most common cause?


Just about everything I find is about GPU reflow, nobody seems to mention RAM.

[stj]

i repair ram faults on arcade stuff all the time - it's probably the biggest failure point.

your problem is a lack of diagnostic software to ram-check the vram.

[Agent24]

Originally Posted by stj

your problem is a lack of diagnostic software to ram-check the vram.

Good point - and even if you find which address(es) are bad, which physical IC does that correspond to anyway... no schematics for these things.

Perhaps freezer spray and hot air could narrow it down - or in the case of such an old card - it's only got 4 ICs - could just replace them one after the other until it works!

[petehall347]

test with another card .if it works leave it be .

[ChaosLegionnaire]

it probably isnt possible to find out which one it is and very hard to test. because all video cards use a crossbar memory controller or basically a raid 0 data striping with the video ram to increase the throughput. the data is striped across all the memory modules on the card so its hard to identify exactly which one or more are failing. im afraid brute force like u said is the only way; keep replacing until it works.

[Agent24]

Originally Posted by ChaosLegionnaire

all video cards use a crossbar memory controller or basically a raid 0 data striping with the video ram to increase the throughput. the data is striped across all the memory modules on the card so its hard to identify exactly which one or more are failing.

Thanks, I didn't know that - I can see it would be even trickier than I first thought... unless something like freezer spray has any obvious effects.

I'll still try that first, and if it's inconclusive, take the brute force approach. I know it's all rather silly for a card this old, but hey, why not?

[ChaosLegionnaire]

another thing further i want to add is that ram is made up of microscopic capacitors, basically... the rule on this forum here is to pre-emptively replace all known bad caps brands even if they appear good.

so some ways to approach and fix the problem are:

if u need to get the video card working again quickly, just desolder all the ram chips with a bga rework station or whatever appropriate equipment and then solder on new/good known working ones to get the video card working again quickly.

but if u have lots of time to kill while waiting for other stuff to complete or arrive, i'd try the brute force approach and replace one at a time to see which ram chips are faulty but personally, i wouldnt reuse them even if they test good. whatever killed one ram chip could have also damaged the others. personally, i'd just replace em all to be safe.

that brand of video memory is on that x300 anyway? tier 1 brand name like samsung and hynix or generic brands like elixir, infineon, qimonda etc.?

[stj]

infineon is not generic, it's Fujitsu!

[Agent24]

If I understand right, DRAM is an array of MOSFETs where the "capacitor" is the gate charge. I suppose it makes sense they all might be damaged if the VRM had badcaps, for example.

The RAM is elixir in this card, whatever that is.

Turns out it's actually an X550... But I guess that's irrelevant. Some cheap "Legend" brand. Apparently has Rubycon MBZ in the VRM.

[momaka]

Originally Posted by Agent24

Just about everything I find is about GPU reflow, nobody seems to mention RAM.

Because in 99.9% cases, any kind of graphical anomalies are a result of GPU chip failure and not the RAM. The RAM actually rarely fails. And if it was the RAM, sometimes you can verify that by increasing the voltage going to it, which should make the artifacts disappear or reduce. Lowering the clock slightly may also do that, sometimes (but it depends on what the RAM is rated for and what it is actually running at, and where you are setting it after that.)

Originally Posted by Agent24

I have an old Radeon X300 which has what is considered a classic sign of bad VRAM; lines down the screen.

No, that's one of the many "classic" Radeon failure modes. Had a dozen of Radeon 9700 cards go that way. x300 is more or less a Radeon 9600 denutted core. (In your case, I see the picture you attached shows a card with only 4 memory chips out of 8 possible, so it probably has a 64-bit memory bus.)

Originally Posted by Agent24

I'm thinking of replacing its RAM chips, but wonder if this is actually likely to fix it?

Most likely, no.

Originally Posted by Agent24

Has anyone done so, and if so, did it work?

I've replaced the DDR2 BGA RAM on a GeForce 7600 GT. The original chips were killed from dead Sacon FZ caps - literally the RAM chips were shorted to ground and the RAM buck regulator coil melted.

I used what was supposedly a direct or nearly-direct replacement chips from an Xbox 360 (Samsung chips, same voltage, speed, and package). Not sure if I killed the "new" BGA RAM chips with the heat while removing them from the Xbox 360 or if they were dead to begin with (the 360 mobo was used for BGA machine profile adjusting at a repair shop I used to work at, so it's possible it's been overheated a few times.) In any case, after swapping chips, my card still didn't work... though I would still call it a MAJOR SUCCESS. Why? Because with the old chips, the video card was completely dead and RAM buck regulator smoking. With no RAM chips, the 7600 GT turned on and displayed a screen full of random-colored pixels (No RAM chips - no telling what's going to get displayed. )
With the RAM replaced, I actually got an okay-ish text image on the screen (no artifacts), a 2D image with some artifacts, and 3D mode always crashed. After turning down the RAM speed, some of the artifacts in 2D mode disappeared, but I couldn't make them go away completely.

So in short... yes, you can swap RAM, it's possible! But as weather you would be able to do it successfully or resolve your problems - good chance, no. Also, your video card has the older style non-BGA RAM chips, and those are even less likely to fail compared to BGA stuff.

Originally Posted by Agent24

Could it be simply bad solder joints

Very very unlikely. Unless that card was dropped from some serious height, or flexed/crushed by something huge. But if that was the case, typically you'd see some ceramic caps break off.

Originally Posted by Agent24

... or that one of the several ceramic capacitors next to the RAM chips has gone shorted?

No, if that's the case, either the RAM or the GPU wouldn't get power due to the short, and then the video card wouldn't work at all. Not to mention that the PSU will likely trip OCP/OPP/SC.

Originally Posted by Agent24

Does anyone have experience with repairing RAM faults, and knows the most common cause?

I can't say I have experience in "repairing" RAM. But from what I've noticed, it's not likely to happen. Only possible exception was some of the RAM in the Xbox 360 consoles, which ran way too hot and cooked itself. When that happens, the RAM cells will begin to require higher and higher voltage over time to keep their stored contents.

And that's part of the reason I still think I might have either overheated the "new" RAM on the 7600 GT card above or that it was bad from the Xbox 360 to begin with... because as I increased its Vdd supply voltage (chip was supposedly rated for 1.8V, but the Xbox 360 fed it 2.0-2.1V) from 1.8V to 2V, the artifacts greatly reduced. Further increasing it to 2.1V, 2.2V, and 2.3V helped a bit with each step, but not that much. And 2.3V was the absolutely max limit for those chips, so I didn't go pushing further. Instead I lowered the clocks and managed to reduce the artifacts a bit more, but not by much again. And there's always the phenomenon, where if you lower the RAM clock too much, the RAM will actually become more unstable and show more artifacts.

So in short, increasing the RAM voltage is one possible way to see if the issue is RAM or not - the artifacts/lines should change/reduce a bit. If no change, it's very likely the GPU then.

Originally Posted by stj

infineon is not generic, it's Fujitsu!

+1

Infineon is indeed a big company that makes a lot of legit RAM. They did have a few bad runs during the DDR2 era, especially with RAM supplied to the Xbox 360s consoles. But otherwise, they are fine. I wouldn't nick them completely for a few bad runs. After all, many manufacturers have had at least one goof now and again.

That said, Infineon is now Qimonda, I believe.
Nanya and Elpida are another two legit brands that make pretty decent RAM (often found in OEM machines from HP, Dell, Gateway, and etc.)

[ChaosLegionnaire]

Originally Posted by Agent24

Some cheap "Legend" brand.

judging from the looks of the sticker, it appears this "legend" brand is from the same oem that makes cards for palit and xpertvision. these brands are known for making products for "budget conscious" consumers so its probably a third tier brand name or such like u said.

Originally Posted by Agent24

Apparently has Rubycon MBZ in the VRM.

it looks like that vrm is mainly for the gpu; looks like single phase power for the gpu. i cant seem to find any regulators for the tsop package vram on the card so it means that most likely the vram draws its power direct from the 3.3v rail. this then means the vram could have been damaged from unstable power from the psu if a "low-cost" psu was used in the system...

anyway, to further add on to what momaka said, sometimes the failure mode of the gpu can be misleading. it could be that one or more of the signal lines that connect the gpu to the vram could have been severed due to bad bga or bad solder bumps that connect the gpu die to the substrate of the chip. this can also cause it to look like the memory is failing. remember, the frame buffer data is striped across the memory chips. if one or more of the gpu to vram connection lines are cut, it will certainly look like the memory is failing due to missing bit transfers caused by the severed connection(s).

so u should probably try overvolting the vram like what momaka said to confirm or disprove the diagnosis but im not sure how its possible when the vram is powered direct from the 3.3v rail... maybe find another spot on the card that has 5v on it and pencil a few lines to the vcc legs of the tsop vram chips? im not really sure how its done, so u probably shouldnt be listening to me on this...

also, overvolting the ram chips doesnt tell u which exactly is the ram chip that is failing. it just tells u whether the problem is indeed vram related or not.

[stj]

btw, touch the rams.
oftem, but not always - a bad ram will run hotter than it should.

[momaka]

Originally Posted by ChaosLegionnaire

it looks like that vrm is mainly for the gpu; looks like single phase power for the gpu. i cant seem to find any regulators for the tsop package vram on the card so it means that most likely the vram draws its power direct from the 3.3v rail. this then means the vram could have been damaged from unstable power from the psu if a "low-cost" psu was used in the system...

The days of 3.3V-powered RAM are long gone. Also, cards that use DDR RAM typically feed it 2.5-2.6V. So there should be a voltage regulator on the board.

Speaking of which...
U3 and U4 are the MOSFETs responsible for generating the GPU V_core rail. (As to why the board manufacturer used "U" prefixes for MOSFETs instead of "Q" - that's beyond my logic. ) U3 and U4 form a synchronous buck regulator (1-phase... which is more than fine for up to 30-60W loads if done right) and it appears the bottom two Rubycon caps (relative to the PCI-E slot) is what filters it. The top Rubycon cap is connected to a different TO-252 device (MOSFET?) on the back of the board, and I imagine this is what generates the RAM Vdd voltage. If that TO-252 device is not a linear regulator and is indeed a MOSFET, then most likely it is driven from a 358 op-amp or similar. Should be pretty easy to trace the "Gate" pin (if it's a MOSFET) to what IC it goes to.

There are also 3 other 1117 linear regulators on the board. One should be for the RAM Vtt rail (which is typically half of Vdd... so 1.25V for 2.5V DDR RAM), another is probably a GPU secondary / bus-level power (so probably 1.2V since the video card is PCI-E?), and the last one is likely 5V for some 74__xx TTL logic (which I think is for the analog/VGA D-sub output).

Originally Posted by ChaosLegionnaire

it could be that one or more of the signal lines that connect the gpu to the vram could have been severed due to bad bga or bad solder bumps that connect the gpu die to the substrate of the chip. this can also cause it to look like the memory is failing.

Exactly!

Originally Posted by ChaosLegionnaire

so u should probably try overvolting the vram like what momaka said to confirm or disprove the diagnosis

And if you do try that and you RAM is DDR type using 2.5-2.6V.... don't go over 2.75V! In fact, even 2.7V could be too much for some DDR chips. See if you can find a datasheet for the RAM ICs and they should tell you the absolute maximum ratings there (or post the memory chip markings, as I cannot read them from the pictures, other than knowing the RAM is made by Elixir... which is an okay brand.)