AMD 3870 VRAM problem?

[cpt.charlie]

Some moths ago i bought an old radeon 3870 for my retro pc project, i tested it for 20 minutes with futurmark and everything was OK so i dedided to put it in storage while i was gathering the rest of the components (dual core 939 platform).
This week i finally got the rest of the parts and started to assemble the pc, but first i had to do a proper cleaning of the card and new thermal paste had to be applied, after assembling it, the card seemed to work fine but ramdomly after maybe 7-10 minutes artifacts appeared on screen (some vertical lines but no BSOD or crashes) if i move a window affected by these lines to a working area, the deffect persist only in the area of the affected window.
So i thought it was a problem with the VRAM framebuffer, maybe a cracked solder joint caused by the heatsink pressure (first time i have broken a card while cleaning it), therefore i decided to reflow the vram chips, i placed the card in a preheater at 130ºC, covered the VRM and GPU chip with foil tape and applied FLUX to the VRAM chips (kingbo rma218), i carefully reflowed chip by chip with the hot air station with no nozzle, air at 50% and 380ºC, i made sure the solder melted by carefully poking with a tweezer on each chip, after leting it to cool down and reassemblig the card again, this time more artifact showed up.

At this point my theories are:

A) The flux residue is causing a defective behavior
B) I damaged some chips with the heat
C) Maybe the problem was the GPU chip
D) Poking around with melted solder balls isn't a good idea (i was very carefully, all chips seem to be in place and varely moved them, so this posibillity is unlikely i think)

My options are:

A) Reball the ram chips (i have never done this before)
B) Replace the video chips (K4U52324QE-BC08)

¿Any advice/suggestions?

Thanks in advance.

[momaka]

Re: AMD 3870 VRAM problem?

Artifacts is almost always GPU fault.
And by that, it's probably not a BGA fault between the GPU chip and the PCB, but rather the GPU chip die and GPU chip substrate.
This may or may not be fixed by a reflow (but is usually worth a try, if you have nothing else to try.)
With that said, there is no point in using flux. I stopped using flux a long time ago, and there has been zero difference in my reflow success rate. Flux is needed only if putting on a new chip or if there is a legit reason the BGA could be bad (like, for example, on LGA 775 motherboards with warped PCB from the coolers.)

As for the RAM chips... I have not see a RAM chip go bad on a GPU - not unless killed by bad VRM/MOSFET or severely overheated.

So now that you reflowed the RAM chips and you have more artifacts, I'm not sure what could be the issue. But nevertheless, first I would suggest to check all of the black SMD resistors around the RAM chips - make sure none were moved from the reflow or cracked when you removed the heatsink by accident. If no issue are found with those, try reflowing the GPU chip.

And regardless if you get this HD3870 to work or not or you buy another one, just keep this in mind:
ATI/AMD was way too conservative with the fan profiles on the HD3870 and HD4850/4870 video cards. As such, these cards often would have to hit at least 70-80 Celsius before the fan started kicking this. This is the major reason why they die / get artifacts.

For the HD4850/4870 cards, I use RBE (Radeon BIOS Editor) and flash my own custom fan profiles so that the fan(s) run at least 30% speed (typically at 35-40C, depending on the heatsink used) and linearly ramp up to 100% at 65C. This does make the cards very loud, but at least they cool well.

For the HD3870... these actually have adequate coolers and will not run as loud. However, I haven't been able to flash them with ATIFlash/ATIWinFlash for some reason. So instead, I've been using MSI Afterburner with a custom fan profile to keep my HD3870 cards cool. I set up the fan profile in a similar way to the HD4850/4870 cards. So even under full 100% load in a hot room (29-30C), my GPU temperatures still stay under 60C most of the time. That's a stark difference between the stock cooling profile, which ramps up the fans only when they reach a toasty 80C or so.

[cpt.charlie]

Re: AMD 3870 VRAM problem?

Thanks Momaka, i completely agree about the high working temperatures of the 3000 and 4000 series of ATI, during the futurmark test i saw this card reaching nearly 90ºC quite impresive.

With regards the flux, i use it to prevent and remove oxidation if possible and at the same time it makes the solder to flow easily as it transfers some heat. I had some hope it was related to the VRAM as a faulty GPU core is a dead card (yep, these chips were affected by the bumps problems), i got some info about artifacts at this link:

http://www.playtool.com/pages/artifacts/artifacts.html

It makes sense but it's hard to know for sure the origin of the problem as the vram is connected to the GPU and a bad solder on the GPU could have a similar effect, i've seen some videos of russian tech fixing artifacting video cards by simply replacing the VRAM chips, but again i don't understand how they can be sure it's caused by the GDDR and not by the GPU core.

I'll try a reflow to the GPU core and post my findings.

[piernov]

Re: AMD 3870 VRAM problem?

Definitely sounds like a dead GPU.

Incidentally I have a 3870 x2 with the secondary GPU artifacting and crashing depending on the application… Main GPU still works fine funnily enough. You'd expect the main GPU to die first, but it may have to do with the cooling system: single blower over the 2 GPU, one with aluminum heatsink the other with copper heatsink. Before repasting it it was going over 90°C in matter of seconds so I didn't check anything else. Went ahead, repasted everything, cleaned heatsink, replaced thermal pads… well now GPUs seem to hover around 90°C even after a long time, but quite sad that the secondary GPU is dead.

Will try to heat the GPU later to see if it makes any difference, I don't think it will though. I usually do 200°C for a minute or two btw, never actually reflowing as there is more chance of destroying the balls than anything. Obviously it won't fix anything, maybe just resurrect it for a few hours/days.

The first graphics card I bought myself was an HD4850 (paid 120€ back in 2008, interesting how it was very cheap compared to middle/high-end cards nowadays), it gave up around 6 years later. IIRC it was hitting 85°C under full load, even with fan at 100%, I wasn't using it that much towards the end of its life.

In any case those never were really reliable…

Sorry for kinda hijacking your thread but I thought this would be relevant to the discussion.

[cpt.charlie]

Re: AMD 3870 VRAM problem?

Interestingly i have another 3870 that came with the now damaged unit, but the working one is pasively cooled, in fact i expected this unit to fail as i expected higher working temperatures on a pasive VGA however it seems to be working just fine, i should replace the thermal paste, but after what happened to the first unit it's better to not mess with "temperamental" GPU chips.

I also have a 7900gt with the same issue, maybe it needs a reflow too, these chips were affected by the bump issue too.

[momaka]

Re: AMD 3870 VRAM problem?

With regards the flux, i use it to prevent and remove oxidation if possible and at the same time it makes the solder to flow easily as it transfers some heat.

But there's no point, because the flux doesn't go where it needs to.

Like I mentioned above, it's not the BGA between the GPU chip and PCB that fails, it's the BGA between the GPU die and GPU substrate. You can't put flux there, because the GPU die has epoxy underfill (the black hard stuff you see around the square part of the GPU die.) This prevents the GPU die from moving after the GPU chips is made (or so it's supposed to be - part of the nVidia "bumpgate" issue is that this underfill goes soft and lets the GPU die actually detach/move itself around and loose contact with the GPU substrate.) This is why reflowing also doesn't always work - the underfill keeps the GPU die from moving / sinking, and hence its BGA balls melting and re-attaching again.

However, the big changes in temperature from the reflow can sometimes cause enough flexing and force to build up between the GPU substrate and GPU die that the BGA balls between the two re-attach. This is when you get a "successful" reflow. Of course, I've also had instances where the same forces have cause partial or complete short-circuit of the GPU main rails.

So with fewer words: reflow is down to luck. It may work, or it may not. It's worth a try if the video card is dead and you have no better options to try to revive it. Finding a new GPU chip would be the proper fix to get it going... but those are usually hard to find or more expensive that the card is worth.

It makes sense but it's hard to know for sure the origin of the problem as the vram is connected to the GPU and a bad solder on the GPU could have a similar effect, i've seen some videos of russian tech fixing artifacting video cards by simply replacing the VRAM chips, but again i don't understand how they can be sure it's caused by the GDDR and not by the GPU core.

Exactly.
I've seen such videos, and it always made me wonder if the heating of the card to remove the RAM chips (and again for installing the new ones) is what fixed the GPU in the first place.

What would be a better test for those videos is to move the "bad" RAM chips to a good-working identical video card to see if the problem moves with the RAM and is exactly the same. If yes, then that confirms the RAM chips are bad. But if not (and I suspect that would be the case), then you know something else is going on.

In other words, if you can't replicate the failure with the "faulty" component(s) on identical hardware, then your "faulty" components may not actually be faulty.

Incidentally I have a 3870 x2 with the secondary GPU artifacting and crashing depending on the application… Main GPU still works fine funnily enough. You'd expect the main GPU to die first, but it may have to do with the cooling system: single blower over the 2 GPU, one with aluminum heatsink the other with copper heatsink.

From what I've seen on most of those X2 cards, it's almost always the 2nd GPU that dies mainly due to the cooling system, as you noted.

In particular, most of these have a blower fan that blows air across the two GPU heatsinks, first through GPU 1, and then through GPU 2. GPU 1 gets the nice, fresh, cool air. GPU 2 gets the already heated air from GPU 1, and thus can't get cooled as well.

Not all ATI/AMD X2 cards are like, but at least the ones with the blower fans seem to be.

On a related note, I recently acquired a dead Sapphire Radeon HD4850 X2. This one has individual heatsinks and fans for each GPU. But the heatsinks are super tiny - maybe good for 40 Watts TDP each at most. A very under-clocked and under-volted RV770 GPU (HD4800 series) will still dissipate around 80 Watts fully loaded. So, it's not surprise that GPU failed at all. I haven't tried reflowing that one yet, but it's coming up on my to-do list, soon.

I tried an HD4870 yesterday, and it was still a no-go... though that one may have other issues.

Will try to heat the GPU later to see if it makes any difference, I don't think it will though. I usually do 200°C for a minute or two btw, never actually reflowing as there is more chance of destroying the balls than anything. Obviously it won't fix anything, maybe just resurrect it for a few hours/days.

200°C won't do anything for the ATI cards of that generation... but give it a try and let us know how it goes.

From my personal experience with the HD4850/4870 cards, none of them came back to life until I really cranked the temperature up. Usually 220°C minimum for at least a minute. On most that I fixed, I had to heat them until the PCB was just about to start to crackle/popcorn from too much heat.

Unlike nVidia cards with the bumpgate issue, where not much heat is required to bring them back to life (or kill them completely), ATI/AMD cards from the HD3k and 4k era will take on a TON of heat before anything happens.

The first graphics card I bought myself was an HD4850 (paid 120€ back in 2008, interesting how it was very cheap compared to middle/high-end cards nowadays), it gave up around 6 years later. IIRC it was hitting 85°C under full load, even with fan at 100%, I wasn't using it that much towards the end of its life.

Yup, the stock/reference cooler on the HD4850 is pathetic.... or rather just inadequate for the TDP of the HD4850 (110W). Actually, if anything, it's a really nice cooler with a neat heat-chamber technology. Problem is, the small 60 mm fan and small surface area of the fins limit it to handling maybe 60-70 Watts comfortably.

Unfortunately back then, not many games offered the option for frame-limiting. A trick to save your GPU from hitting high loads is to limit the game FPS to a value that the GPU can handle comfortably without hitting full load. I do that on all of my current GPUs right now to keep the load at 50-70% max (rarely peaking more.) This allows me to use most of my high-end cards with their stock coolers and still keep the GPU easily under 55C.

but the working one is pasively cooled, in fact i expected this unit to fail as i expected higher working temperatures on a pasive VGA however it seems to be working just fine, i should replace the thermal paste, but after what happened to the first unit it's better to not mess with "temperamental" GPU chips.

Passive-cooled HD3870?
That mush have a pretty big heatsink then.

I love passively-cooled video cards! - that is, after I put a large fan on them. Even if the fan is not so large and doesn't move a lot of air, the difference between some air movement and none is still grand. Even a tiny air movement often improves the temperatures drastically (on top of that, without any significant noise.)

I also have a 7900gt with the same issue, maybe it needs a reflow too, these chips were affected by the bump issue too.

Good luck.
I have not been able to revive an nVidia bumpgated card with a reflow up to date. Most ended up just short-circuiting on me, even if I was very careful with the heat.

On the other hand, I've had decent luck with nForce 6100/6150/6200 MCP/chipsets.

[cpt.charlie]

Re: AMD 3870 VRAM problem?

About the flux i know it won't affect to the internal micro BGA, but remember, i was reflowing the VRAM chips.

This made me think about a crazy idea to test one day, maybe remove a bit of the underfill in oposite sides as it just cover the edges, inject some liquid flux and reflow the core. I guess it uses higher temperature melting point alloy so maybe 250/260ºC needed, if it works reseal the chip with epoxy, the problem of this plan is that underfill is quite resistant and very hard to remove without damages on the core or substrate, as i said it's a crazy idea.

I love passively-cooled video cards! - that is, after I put a large fan on them. Even if the fan is not so large and doesn't move a lot of air, the difference between some air movement and none is still grand. Even a tiny air movement often improves the temperatures drastically (on top of that, without any significant noise.)

It's quite big as you can see in the photos i'm not sure if i should add a fan and connect it to an available fan channel of my scythe kaze server or take the risk of replacing the thermal paste.

[momaka]

Re: AMD 3870 VRAM problem?

This made me think about a crazy idea to test one day, maybe remove a bit of the underfill in oposite sides as it just cover the edges, inject some liquid flux and reflow the core. I guess it uses higher temperature melting point alloy so maybe 250/260ºC needed, if it works reseal the chip with epoxy, the problem of this plan is that underfill is quite resistant and very hard to remove without damages on the core or substrate, as i said it's a crazy idea.

Yeah, I've had a similar idea before too.
Don't think it's going to work.
That's not to say it's a bad idea. Sometimes you never know until you actually try.

It's quite big as you can see in the photos i'm not sure if i should add a fan and connect it to an available fan channel of my scythe kaze server or take the risk of replacing the thermal paste.

Looks similar to the heatsink on my Radeon HD7950... except mine has a fan on it. Of course, the HD7950 is rated for up to 200 Watts TDP, while the HD3870 is 106W. But with no fan and just relying on air movement in the case... I think the cooling capacity of that heatsink will be severely reduced.

You should put a 92 or 120 mm fan on it... or a pair of 80 mm fans... and run them at 5-7V constant. This will keep them virtually silent while greatly improving the performance of the cooler. Probably won't see more than 60C under full load. Or use a controlled fan header that bumps the speed up with GPU load, and you may get even below 50C under full load with a little higher fan speeds.

[goodpsusearch]

Re: AMD 3870 VRAM problem?

I always put a fan on any passively cooled GPU. It makes a very big difference and actually is enough for the card to stay cool under any circumstance.

[cpt.charlie]

Re: AMD 3870 VRAM problem?

Here is an example of what i said about the memory, it seems to be damaged due high working temperatures in this case, but no visual damages or shortcircuit detected.

https://www.youtube.com/watch?v=a5sw61l1eJk

[cpt.charlie]

Re: AMD 3870 VRAM problem?

Today i had some time and i decided to reflow the card using the attached profile in my DIY reballing machine but unfortunately it didn't have any affect at all, it's a pitty, i guess at least i have a nice paperweight

[ChaosLegionnaire]

Re: AMD 3870 VRAM problem?

u probably need higher temperatures to reflow the solder bumps between the gpu die and substrate but the risk is popcorning the gpu like momaka said.

[momaka]

Re: AMD 3870 VRAM problem?

Here is an example of what i said about the memory, it seems to be damaged due high working temperatures in this case, but no visual damages or shortcircuit detected.

https://www.youtube.com/watch?v=a5sw61l1eJk

Once more, I don't like to be a nay-sayer, but videos like this aren't necessarily proof of anything.

Again...

What would be a better test for those videos is to move the "bad" RAM chips to a good-working identical video card to see if the problem moves with the RAM and is exactly the same. If yes, then that confirms the RAM chips are bad. But if not (and I suspect that would be the case), then you know something else is going on.

In other words, if you can't replicate the failure with the "faulty" component(s) on identical hardware, then your "faulty" components may not actually be faulty.

I know it would be a lot more effort for the guy to take the dead RAM chips and put them on a good GTX 1660 GPU. But that is the only way to show those RAM chips were actually bad and it wasn't just a coincidence where heating the whole card (particularly the GPU chip) while replacing the RAM chips fixed it. It's understandable that GTX 1660 cards are still worth quite a bit of money, so I can see why the guy isn't taking another working one and putting the bad chips on it to show the issue. Problem is, a lot of people make money off of YT channels/videos. So for any sort of credibility, I think it's important to show this. Again, not saying this YT channel in particular is posting fake stuff... but with so much fake crap and bad info out there on YT, I simply no longer trust the majority of stuff - at least when it comes to technical info and repairs. When it comes to that, I think specialized forums are still a bit better in most cases.

u probably need higher temperatures to reflow the solder bumps between the gpu die and substrate but the risk is popcorning the gpu like momaka said.

IDK, looking at the results of the temperature profiles that cpt.charlie posted, it seems the GPU chip should have reached proper reflow temperatures.

So maybe it's just dead. Or maybe multiple reflows could get it to finally work.
...
Or yet another maybe, perhaps cpt.charlie turns out right and the RAM chips are the cause of the issue. Unless the incentive is purely to satisfy curiosity, though, it doesn't look like any further troubleshooting will be attempted. Though if you are still curious, cpt.charlie, let me know. I do have a HIS Radeon HD3870 that appears to have a dead short-circuit on the RAM Vdd power rail, and it's not the RAM. My HIS card (lol, that sounded funny) is basically a reference card. What's neat about the reference card is you can remove some jumpers to separate the RAM Vdd into two parts - the one that goes to the RAM chips and another that goes to the GPU. The one that is shorted on mine is the one that goes to the GPU chip. So it may be a bad/shorted ceramic cap somewhere. However, I suspect shorted GPU. RAM chips read normal resistance, though. So seeing that you have the tools to remove/reflow/reball BGA, if you do want another HD3870 PCB to play with, let me know.

[Deusjevoo]

Re: AMD 3870 VRAM problem?

Today i had some time and i decided to reflow the card using the attached profile in my DIY reballing machine but unfortunately it didn't have any affect at all, it's a pitty, i guess at least i have a nice paperweight

Your profile is ok, but I'd go a bit slower.

Do you know the history of that card? Has it been reflowed maybe before you bought it?

If a BGA fails it's mostly the solder. Mostly, not saying it'll never be the bumps but that's rather an exception. Was more a problem during transition from lead to lead free period probably.

Reflowing MIGHT fix it but although it was still working that card probably already had bad solder connections. It'll only get worse over time.
Every heat cycle can add to it. Not only starting and shuting down but also the varities while working, iddling, working...
Your last test might have "cracked" it after cooling down.

Reflowing can even make it even worse instead of fixing it as heat = oxidizing.
Solder balls and/or -pads can be oxidized, still touching each other but not making electrical contact anymore due that oxydation. Air = oxygen might still reach it also making it worse when shelving it a long time.
Possible the flux won't get under it to remove the oxydation meaning reflowing something that doesn't take solder anymore = fail.

I'd reball it. Most BGAs fail on solder not internal.
If not fixed, can still try the memory later.