New love for old cards - [GPU restoration]

[Dinnercore]

Another update, did you miss her yet?

Oh she is a demanding lady. Remember how she passed Furmark? Yeah, but as soon as I tried any game I ran into trouble. First wanted to go for Sacred 2 and it only took 20 seconds for the menu to freeze up, come back with stutter and then freeze for good. Okaaay....
Well. Could be driver and SLI being stupid. Fired up an older title (Gothic, which ran fine on the same driver and the 9800GTX+) and deactivated SLI. This time it lasted 3 minutes and ended in a hard lock-up with me having to reset the pc. This does not look good.

I now had 3 options instantly in my head. First, driver still wonky. Not very likely after it ran Furmark in SLI without a hickup. Second, well I was not feeling confident about those 0.5mm pads.... Third, this card has trouble beyond my grasp, since Furmark did not stress memory, maybe I have a bad solder joint on a memory chip or it degraded for good.

I´m not having any of option three, thanks. I will test option one, but first I have to take a look at those pads and make sure they contact.

Atleast the paste looked decent. Sad I had to open it up again. I´m getting quicker in taking these cards apart.

So I looked at the pads, and forgot to take a picture because I finally wanted to move on. Well they did not really connect well with the plate. They had some loose contact and looking at them I saw that varying from 20% - 70% of the surfaces had left their mark on the plate. The SLI chipsets looked very good. These pads took most of the mounting pressure and were a quite dented in. So I decided, just put 1mm thickness all around and it should work. Absolutly perfect would be something like 0.9mm on the memory, 1mm on the chipsets and 0.8mm on the backside memory. But I only have 0.5 / 1 and 1.5.

Looking a lot better then before, and it felt a lot better while tightening the screws down. I could feel the soft resistance of the pads this time.

Now I did not waste the other thermal pads I removed, I put them flat on a piece of plastic and put them in a sealed container for later use. Sorry to bother you all so much with this one, but I want to do things right and I stand by my mistakes. Now I just have to ask a mod for permission to edit my earlier post so no one who googles this gets the idea that they can order 0.5mm pads for this card!

Man it looks so beaten up and scratched. Reminds me of my old skateboard.

But she works now. I did not want to go full force today and went for an hour in Flatout 2. Finally. Fingers crossed.

Our next candidate will be from team red, before I get a bad reputation as an nvidia shill.

An all-in-wonder multimedia solution. GPU and TV-card in one. Set in someones basement for years until I took it out of it´s premature grave. But this is a story that I will get into later. The 295 really clings to me.

 

[droopyRO]

Iirc those cards were loud.

 

[hat]

Maybe not after Dinnercore works his magic. Clean it up, fresh paste...

But damn, where have all the single slot cards gone? I'm sure a single slot cooler would work at least up to GTX1050Ti...

 

[Dinnercore]

Iirc those cards were loud.

Oh I heard that now. If that thing spins up to 100% you hear it. Not just in the room the PC is at. It´s not just that I measured 54dB next to the open case, the high pitch of that fan is not pleasant.

Maybe not after Dinnercore works his magic. Clean it up, fresh paste...

But damn, where have all the single slot cards gone? I'm sure a single slot cooler would work at least up to GTX1050Ti...

Heh, I could do a lot of things to silence that cooler, but that goes a bit against my intentions of restoring them. If it comes with a stock cooler I want to do everything I can to make that stock cooler work like it did at day 1. And those were just loud. Cleaning it up will not help this one much I´m afraid. We will see. I hit another brick wall anyway...
I think the noise is one of the main reasons those single slot solutions are gone. At that low profile point you can´t waste any cooler mass and a radial-fan is pretty much your best option there. However to make it work with fins that can´t have any siginificant height, you need high rpm = a lot of high pitched whiny noise.



Well next update: Bad news. I think my MSI-X370 Gaming Pro Carbon already gave up on me. The thing they advertise with PCIe 'Armor' to withstand large GPUs and multiple swaps. Ye, everything from marketing is always a lie. If they have to say it, it can´t be true.

I did my usual routine, use DDU, shut down, unplug power, take out previous GPU, plug new one in. Boot back up, so far so good. Got into Windows, and installed the legacy driver I found for this card on the AMD-Website. Then I just had to restart....
It started with the issue that the PC would not restart, e.g. not shut down. It got to the point of signing out of Windows only to stop there with peripherals already powered off. And from here it just refuses to post. The X1900 screams with 100% fanspeed, everything powers on except for RGB-strips. Debug LED on the board show nothing, it just stays in this state forever. Waited 30 minutes, nothing. Double checked every power connector, checked my AiO for a leak, everything ok. Tried another GPU that worked before, nope.
It just won´t post, and doesn´t even get to any debug LED state. No Idea what killed it.
Maybe my 1.40V OC on the CPU was to much, maybe the PCIe slot died. I now have to fun task of troubleshooting my fairly new motherboard, of course the part that is not 10+ years old craps itself when being stressed.

I´ll clear CMOS, use another PCIe slot, switch PSU, I guess you all know the drill. But project is on hold until I sort it out. I have an LGA775 board ready, but nothing to use it with yet.

 

[hat]

I never had an issue with "loud" stock coolers, though. Sure they'll get loud at 100% (most fans do) but it shouldn't reach 100% anyway.

Sorry to hear about your board, though. It's odd that it would just die on reboot. I've killed an AGP slot before, and the system would have run fine otherwise, just no AGP... so I feel like something else is amiss. Not sure what could have possibly happened though, swapping GPUs shouldn't kill a motherboard, unless there was a mighty ESD.

 

[theFOoL]

By my Past use with these "Single-Slot Cards" (Please bring back God ) The Fans would go to 70-80% or even 90%

 

[Dinnercore]

I never had an issue with "loud" stock coolers, though. Sure they'll get loud at 100% (most fans do) but it shouldn't reach 100% anyway.

Sorry to hear about your board, though. It's odd that it would just die on reboot. I've killed an AGP slot before, and the system would have run fine otherwise, just no AGP... so I feel like something else is amiss. Not sure what could have possibly happened though, swapping GPUs shouldn't kill a motherboard, unless there was a mighty ESD.

That one time the card booted up for driver installation the fan did drop to idle and it was ok in that state. Not louder then anything else in my system.

I´m a bit puzzled too about this issue. What on earth can happen on a reboot that the whole board just bricks. I watched it closely and I do get a confirmation on the RAM-Slots from the board, they work and get detected. I do get the CPU-light and on the debug thing it checks CPU and passes that, but as soon as it should move over to GPU it just stops. No light nothing. But the PCIe slot LED is on, confirming that the board has detected something is plugged in there. I dusted everything off now, tried the 2nd slot but nothing so far.
I tried to switch the PSU cable, I tried a different plug on the PSU for PCIe power. I decided to just leave the power connector unconnected on the card and just try to get past GPU detection but nope.

I unplugged all drives except boot drive, nope. I confirmed the AiO pump is running. No leak.
I had power off completly and cleared CMOS, nope. I tried brute force on/off/2x reset, nope. Something is not quite right. I remember having the same issue before on the 9800 and once on the 295, but everytime resolving it by switching PSU cable from using 2 seperate to a single one. On the 295 the other way around... But now this no longer works.
Could be the PSU but wouldn´t it be more likely that the whole 12V supply would be faulty then and not just the PCI connectors? Ahh man.

 

[Mr.Scott]

Could be the PSU but wouldn´t it be more likely that the whole 12V supply would be faulty then and not just the PCI connectors? Ahh man.

Depends on how many 12v rails it has.
If it's not a single rail PSU, the PCIe connectors are usually on their own rail. So it's entirely possible for just one rail to be bad.

 

[Dinnercore]

One for sorrow // Two for mirth, // Three for a funeral // And four for birth.

System is working again. Bios was the issue. Had taken it apart (and gave it a proper clean-up in the process) and rebuild the whole thing. Was still not working, so put my tweezers between the battery jumper pins for a good 10 minutes instead of just 10 seconds and guess what, it´s as if nothing ever happened. MSI please learn how to bios, it takes more then talent.

Now for that lovely all-in-wonder card.

The most interesting feature set out of all the ones I have so far. Performance is not it´s strength, but it was intended to be a multimedia powerhouse back in the day and oh boy this card can do things that no other solution can. Came bundled with software for DVD playback and creation. Well not a big deal. But you can hook up cable TV to it, PAL/SECAM and it supported DVB-T. It has it´s own tv-tuner and a co-processor 'Theater-200' for video and audio De-Interlacing. Hardware accel for H.264 and you can hook up a radio antenna and use it to listen to the radio over your pc back in the day!
Want to watch TV while working? Well you can set the TV-Screen as your bloody desktop background!
It even had a crude time-shift function. I´m astonished by this thing.

Very dirty on the upper side, but the back looks clean. Must have been sitting in an open box face up.

Getting it to work was simple (only MSI-bios messed up), AMD still provides the legacy ATI-driver software. Thank you AMD!

Right from the start the idle temp had me worried. The fan was chilling out tho. Ambient this time 21.5 °C.

A bit warm but they were designed with that in mind. Instead of making noise with that screaming fan they let the card heat up. I don´t agree with that completly, now that it aged and electronics do degrade faster with heat, but eh.

I wanted to see if I can control the fan with afterburner and noticed an interesting thing:

It picks up 2 processors, the R580 GPU and something unlabled. No fan-control for both, but I think that 2nd could be the theater-200.
Since it idled that high I didn´t want to throw Furmark at it instantly. So some medium load first.

Guild Wars should run fine on this. But 70°C and the fan is still not bothered. Well, if they don´t let the fan spin anyway I don´t hope to achieve much with my fresh TIM. A soaked heatsink is a soaked heatsink, I will only get it to soak faster.

Time for Furmark:
Ambient still 21.5°C

It does get hot, but not because that fan and heatsink can´t handle it, but because they didn´t let the fan spin up to keep the noise low. It was audible at around 40% but not really a bother. However it did drop back down to 36% at the end of the test, because it cooled down to 75°C on 40% fan speed.
From what I see this thing has a temperature soft-target of 80°C and tries to get there and hold that with the least spins possible on the fan. I´m not okay with 80°C, looking at how easy the temp went down to the low 70s on just 40% fan-speed but I can´t control it atm.

Power consumption idle: 125W // load: 175W.
Temps idle: 52.8 °C // load 80.6 °C

Now taking it apart:

Looking good. Only had one screw on the cooler that wouldn´t turn and I rounded the head. Got it out by careful use of pliers and will find a replacement for it.

I had to look closely at the red comb to get it off, you can pull out plastic pins in those knobs that hold it in place and then squeeze them to get them out:

That´s it for today, next update will be cleaning and close-up pcb shots with the new thermals and testing afterwards.

 

[theFOoL]

That's it for today, next update will be cleaning and close-up pcb shots with the new thermals and testing afterwards.

Take off the white thermal string so can see those things

What are those white thermal strings for BTW?

 

[hat]

VRMs?

If you really want to tweak that card in particular, try out TPU's very own ATiTool. It's pretty well antiquated now, but it should work for that card. Or maybe you can edit the BIOS with a higher default fan speed, but modding the BIOS probably goes against the whole restoring vintage stuff thing.

 

[basco]

and i think gputool loves the old voltcontrollers too-its worth a shot
https://www.techpowerup.com/download/gputool-community-technology-preview/

and is atitool 27 beta 4 the latest?

 

[dorsetknob]

Now for that lovely all-in-wonder card.

great job

Being a Fan of the AIW Series of ATI Graphics Card its a shame they Stopped making them
I Started off with a 3DFX 3500TV (agp2)
Forced to upgrade because it could not handle Agp4 or 8 i looked at NV offerings (Crap) and settled for the 9800se AIW
Great Card lasted years (bugger is still working but boxed for storage)
Only gave them up because those that supported DVT were not available locally and mine was analog only.
ended up with a pci TV Card.( huppagage).

Still look for available AIW Cards on Fleabay ( their in short supply for the later models )

 

[hat]

These days any card can do what an AIW did. Just run an HDMI to your TV. Back in the day though, they sure were useful for connecting a PC to a TV.

 

[Dinnercore]

These days any card can do what an AIW did. Just run an HDMI to your TV. Back in the day though, they sure were useful for connecting a PC to a TV.

Those were not just for connecting a TV to a PC like you suggest. It does more then any new card can, it makes your pc into a TV. With the build in tuner. Or can you use your 1080 to watch 2 tv channels on your pc while browsing the internet and having a 3rd channel record for time-shift playback? I know no card that can take a direct coax-cable and the aiw can take 2. And I can´t listen to the radio over my Vega 64. Sadly.

If I run an HDMI from my tv to my pc, why use the pc in the first place? With the aiw I can save the cost of the TV.

 

[Caring1]

I used to have an A.I.W. in a home theatre box in the loungeroom just for that purpose, sadly the advent of D.T.V. saw it retired.

 

[Dinnercore]

Hope you didn´t fall into kernel-panic (low freq. warning on that link, please keep volume low at night if you like your neighbours) in the past days. The aiw X1900 is done, the messed up screw was more difficult to replace then I thought. In the end I found a matching one in my big general collection box.

But that is boring stuff, you want to see the freaky VRM under that thermal-pad band?

Look at those sweet colors!

I guess it must be some anodizing but I don´t really know the reason behind that. Maybe there is someone here who knows more about this and why it is colored like it is? Does it have a meaning or is it just due to manufacturing process?

Moving on to the die:

Beautiful R580. I think gold suits it very well:

Then we have the Theater-200.

And the Nxt-6000 as COFDM Decoderchip for the TV-Signal.

Cleaning it up was easy, not much to do. Some wipes, some brush strokes for the dust that sits in tight places. The copper part of the cooler could be taken apart by undoing the plate with the friendly lady and her friendly sword.

Note: I did re-use the pads on the memory, because they looked ok and had no dust on them. And they are 2mm+ thickness, my new pads only go up to 1.5mm. I did however replace the strip from the VRM cooler with some of the leftover 0.5mm ones that did not fit the GTX295. That was a good decision as it turned out!

I did not like the look of that old copper, even tho copper patina has it´s own appeal.

Quick brush under citric acid, very short time to prevent copper citrate.

And it´s all back together!

The results on thermal performance did make me scratch my head this time. With an ambient of 23 °C, we are just barely above the 21.5 °C from the first test. Still I got this:

Idle pcb-temp went up by 5°C, while the GPU-Temp is within the same range as before. The VRM however did drop a full 6 °C with the higher ambient temp. That is some serious improvement from the new pads.

New load temps are interesting too, the fan had to spin at 45% the whole time to maintain the 80°C on GPU-Core. PCB-Temp is now 6°C higher then before but the VRMs are 16°C! cooler and remember the slightly higher ambient.

Not sure what to make of this result, I´m happy for the VRM but worried about the pcb and core-temps. Since the VRM-cooling-comb-thingy has no contact to the main cooler the increase in GPU-temp can´t be from a better transfer. Maybe the old pads on the memory have to cure again, or did not re-shape very well. My paste application was generous and I always spread it manually to cover the whole die.
I will work with this card for a few days now and see if there will be a change.

I wanted to give ATITool a try, but I could not run it because Win7 'saved' me everytime from the missing driver signature for this program and prevented execution. I could not bypass this yet.

 

[hat]

There's a way around that... advanced boot menu or something, should be an option to disable driver signature enforcement.

 

[theFOoL]

That "Disable Driver Enforcement" should be in the Advanced Boot Option which one should be able to Access by pushing the F8key before the boot logo appears

 

[Dinnercore]

I will give that driver enforcement setting a try in the future, thanks. But today I wanted to move on and didn´t fiddle around much longer.

Last thing I did with the AiW X1900 was play some Half Life 2 Lost Coast. I don´t have the full game, but this tech-demo level will do:
(Thank you @jaggerwild for the suggestion)

I like to blow those boats into pieces and watch them float on the water. Stuff breaking apart and floating on water in a game blew my mind as a kid, I imagined all the calculation horsepower it would need to really simulate such detail, ofc I did not have enough understanding and knowledge to see that physics were and are still NOT simulated but rather emulated with shortcuts and simplifications that I consider cheating. But that blissful ignorance back then made it all the more enjoyable. The more you learn and know, the less magic you find in the world...

Anyone want to guess what happend right after?

Average fps were around 35-40, resolution 1080p! Not bad for 256mb memory. And additional stuff like texture filtering was enabled too. Only texture detail on medium and shader on medium.

The reason I wanted to move on was a patient I received today:

7950 gx2. Sold for the equivalent of a loaf of bread and described as not working. Hence all the missing parts, this poor thing was already stripped for the bin. Standoffs are missing, slot-bracket missing.
That´s how it presented in the ER this morning, no further history known.

If it´s sold as not working and someone already gave up on it, I propably don´t have much success either. But I have to try and I like myself a challenge. No burn marks or browned stickers, this thing did atleast not get treated with oven necromancy yet.

Quickly grabbed some standoffs and estimated the length:

Looks alright eh? A bit scratched:

Now how do you support a card in your pc when it does no longer have a slot-bracket? I tried to fit one from an 8800gts, they have the same ports but it is spaced in a different way so I could not get the card to fit with that one.
Well just stuck something under there for now:

And the moment of truth, does it do anything?

Well it is still breathing, I have no idea where to measure pulse on a GPU, but no display out. Red light of doom on mainboard debug LED. Tried bios reset just to be sure, nope the card will not run like this. Funny enough, on the second try it did get stuck again between CPU self test and VGA test, with something detected on the PCIe slot, but it did not boot any further.

Fans both spin on the top and bottom pcb, full 100% like they should. On the pcb with the PCIe connector the VRM is heating up and getting lukewarm, the other pcb I could not feel any heat just the fan spinning. Could be a clue.

I will take it apart now and scan the pcb and SMDs for any visible damage. I really hope I do find some, if not I got a real problem.
Has anyone got a source for pcb blueprints on cards like this one? So I can check where each component should be? Will google myself, but maybe someone knows of a collection somewhere that I don´t come across.

 

[hat]

Seems to me the card was dropped or something. Could be physically broken somewhere...

 

[jaggerwild]

Hard to find a 7950x2 that does work, think heat was an issue not certain. Might contact a gpu seller like Evga they might sponsor a gpu flashback(worth a try)!

 

[Dinnercore]

Seems to me the card was dropped or something. Could be physically broken somewhere...

The one corner of the pcbs closest to memory chips seem a bit scuffed. Could have been dropped, but I could not make out any other physical damage.

Hard to find a 7950x2 that does work, think heat was an issue not certain. Might contact a gpu seller like Evga they might sponsor a gpu flashback(worth a try)!

Yeah I read now that they ran hot all the time with the stacked pcb there are little options for improvement with aftermarket cooling, except special waterblocks designed for this card...
I hope I find one working, and if I do I will not use it until I figured out a way to run these without cooking them. But before I will contact EVGA, I´ll rather buy some more random cards. For the price of shipping to USA I´d get about 3-4 of these cards.


I did spend some hours with the 7950 GX2. There was no dust or anything to clean really. I first visually inspected it as well as I could with my trusty 10x folding magnifier that accompanied me during my geosciences study.
Except for scratches on the cooler I found nothing. Solder joints looked bad in some places, but I could not spot a loose contact.

I did not find specific information on the pcb-layout but some datasheets like the one for the ISL6568 two-phase Buck Controller. So I went ahead and stabbed the thing with my multimeter, first powered off on my desk, since that is the only way I could get to most of the contact points anyway. Can´t check voltages that way, but I could check if everything is still connected.
Well it all seems fine, just to make sure I checked every pin of that connector that bridges the stack together. Nothing wrong.

Took my close-up shots:

The Intersil:

Then put everything back together. Plugged it in one more time, but no. Mainboard reports VGA-Error on debug. Both fans spin, this time both boards got warm. I measured voltages and everything seemed like it should work. 12V supply was stable and on both pcbs. Just for fun I switched the DVI-port but ofc that was useless.
For now I´d say this is beyond what I can repair, since it can not be tackled by replacing components on the pcb other then new cores and/or memory.

However I refuse to ever give up. This stays on my to-do list, but down at the bottom. I will try to get any signal from it with a board that features a chipset that is actually listed as compatible with it. And if this fails too, well I do have a hot-air station. Might try some heat necromancy. And if nothing, well I always wanted to do a crazy core transplant. Why not go full maniac and do a reball. But for now, I´m afraid we have to move on.

Question is, what comes next. Another dual GPU or are you already tired of them?

 

[hat]

How about a PS1?

 

[jaggerwild]

Can you use a long stack to space out the GPU'S from each other and extend the fan cooling wires? Clearly with the short stack the cooling isn't there for the underside gpu. Longer ones will work.