New love for old cards - [GPU restoration]

[Dinnercore]

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.

That doesn´t really work because the cards need to talk to each other, there is only one of them that has the connection to the mainboard. This is where this bridge pcb comes in:

I can´t extend that easy. There was a version that was higher or even adjustable via ribbon cable (not sure) for some aftermarket cooler. But finding these today will be difficult.

 

[Dinnercore]

New Update. I don´t know if I can do this anymore... you open a box that got send to you and unwrap the two 8800 GTX you bought, just to hear that high pitched clacking on your floor of small metal fragments falling out. You know, I have no problem seeing gore on people, but this moment made me a shook one.
The previous owner didnt care about them, told me only one works the other sometimes picks up and sometimes doesnt. Auction ended very cheap so he just careless threw them back on back in the box...
I guess this will happen frequent. Warning, shocking images ahead.

There was some serious force involved to rip the layer off of the pcb and do things like this:

And this is from the better looking one. The other has even more damage. Despite how it looks I did pull out my soldering stuff and tried to get atleast the ones that are just loose back in place. Lead free stuff gave me some trouble to reach the necessary temp, because the kit with the fine tip for small electronics I own was the cheap sidekick to my main one, but that is way to big for these small things. In the end it was no use anyway, both cards definitly dead and I´m missing more SMDs then I could find in the packaging

On the plus side there was some more stuff in that box, I now got 3 more Arctic Accelero aftermarket GPU coolers (unused) and a smaller Alpenföhn Klara (unused too) that fits mid tier cards from ATi and Nvidia (6600 e.g. or the ATi X1900). And a waterblock from aquacomputer for 9800 GTX(+) cards. Oh and another 7950 GX2, this one will be worth a closer look because it does boot, and with the driver I had I got into the windows login screen that was correct 1080p, and then it dropped signal. Went back on and off and on and off in a cycle, indicating overheat. The fans did not spin up at all too, they were spinning but not audible. But that´s for later.

Actually I had this guy planned for a while now, another card I got for cheap as it was sold as 'untested and stored on display for years, can´t test if working'. I guess most people thought 'oh yeah sure you can´t just plug it in your pc and do a quick test boot, it must be broken'.

We will see:

A 9800 GX2, with original box and even the game it came with (Neverwinter Nights 2). Only the outer shell/cover is missing the screws, I guess this card was running without it for thermal improvement and then the screws ended up somewhere else.

Top:

The sideview with the ribbon cable to connect both cards:

And the backside. I had to photograph it standing up because on the top side it has small metal contacts on very thin and bendy metal springs that tend to break off. These must be to route ground to the metal cover.

The fan is covered in dust and since it didn´t run for ages I had instinct telling me to not plug this one in right away and clean it up first. But I gotta do what I gotta do for data. Let´s see if it runs at all before I invest time.

Put it in the slot, flipped the power switch, pressed the botton aaaand. Blackscreen. Nothing. Boot stuck at red-VGA error LED. ARGFDSGSF

Okay. Wait a minute. Card shows green LEDs on power connectors. Fan spinning and not just spinning but ramping down from 100% to idle after starting the pc. This smells like a working card and me being dumb.
Looking at the back of the card I find 2 more LEDs, one for each pcb. I put the monitor DVI-cable into the top port, this one showed green. The bottom one with the empty DVI and HDMI port had a blue LED. So I switched my cable down to where the blue light was and powered off and on again.

Now its working!

There we go, detection works fine.

Horrible Idle temps, but I read in the original review of this one that 60+ °C is the normal idle temp with this cooler since the card has no 2D-downclock mode. It runs at full speed 24/7.
22°C ambient today. 175W power consumption // 61°C on the hottest GPU.

I do not like this. Especially that 3 °C difference, could be a hint that one side has less contact then the other. Well I can always quit the Furmark test if I get uncomfortable. They should shut themself down at 105°C, but I don´t want to heat a card like this up for no reason considering the age of the parts.

And this is where I cut the test short. I think everyone can plot the curve a bit ahead and see that this one climbs steadily into oblivion. 100°C at the 5 minute mark I would guess, and maybe even 105°C before 10 minutes.
Fan was already at the limit. We reached a thirsty 370W max. and went to 92°C on the hot core after just 2 minutes 11 sec.

But I´m happy that it ran stable. Had it working for over an hour in windows and played Flatout 2 for 20 minutes (reaching 72°C max.). No crashes or trouble so far.

Next is teardown.

 

[hat]

Too bad about the 8800GTX... but nice score on the 9800GX2 at least.

 

[Dinnercore]

Sorry for the long wait, could just get back to it this weekend. New semester began, gotta split my time between study and work, leaving not much room for other things.

Well then, we left at the furmark test of the 9800 GX2 and now it´s time to crack it open and look if there is any room for improvement, side A:

Very little dust got in there, looks like the cooler does a decent job at keeping dust away from the pcb. Here is side B:

Paste was still wet in some spots and looked fine. The pads were a bit crumbly but not bad. Not much cleaning to do for me this time either. Not much I can do about the high temps here, seems like the limiting factor is the cooler design itself. Has very large fins, low fin count and is all in all more mass then surface area.

Here are some shots of the silicon:

Looking around the board I found an interesting name on the sockets for the ribbon cable connectors that link the two pcbs together:

Not sure if that is a part from THE Honda automotive company, but might be.

After the photo session I got my MX4 and the Arctic Cooling thermal pads, read online that the pads should be rather thick and put everything back together. Again it was bad to just assume the thickness and I made the same mistake as I did on the GTX295. The correct thickness is again 1mm all around! (I need to order a huge amount of these, they seem to be most commonly used)

This is what the contact of the die looked like:

NOT good. I´m lucky it didn´t go worse. I even tried to check by looking at it sideways from an angle and using a laser pointer to try and see if the die had contact but I couldn´t tell.

So I had it running (the pic is from after I noticed!) and checked temps, seeing that on one GPU it climbed up to 70°C and more in idle, telling me that I made the pad mistake again... Which meant that I had to undo every screw again, and remove the I/O-cover too because in order to seperate the pcbs you need to take that off too, which is a pain because you need to reach the screws through very small holes on one of the boards.
After those pleasant 30 minutes of undoing and re-tightening dozens of tiny spring-loaded screws that love to bounce everywhere, it now is finally done and back together.
Even found three screws for the cover to hold that in place again:

Man this card is looking nice. I like it.

A little damage around the power connector area, adds some character

Now my time is already running out for today, so new temperature figures will come tomorrow. I don´t expect a big gain however, these cards are known for load temps in the mid 90s.

To finish it for today, here is the card in action right now:

 

[Radical_Edward]

This thread inspired me to take apart my old ASUS HD5850.



All cleaned up. There was a full blown floof ball in the heatsink too. The TIM on the chip was hard.

 

[Dinnercore]

This thread inspired me to take apart my old ASUS HD5850.

View attachment 109070View attachment 109071

All cleaned up. There was a full blown floof ball in the heatsink too. The TIM on the chip was hard.

View attachment 109072

Wow that looks like some years worth of floof. Good thing you took it apart, did you put it back together yet?


My next update:

I did no good to this card it seems. First off the new idle temps are kind off ok:

19,7°C intake air today, but the numbers did not change really when compared to last time. The curve in afterburner shows how the temp climbs up after start up.

Atleast it doesn´t exceed 75°C now, like it did with bad die contact and it needs less RPM at the same temp. I expected now from the load test, that I would see something like 90°C again, but not much more and that it would hold there.
BUT:

It just shoots up faster then ever, especially on the one GPU that has the PCIe connector on the pcb. I aborted the test, it did not crash and since it does take a minute to get to 90°C I think the cooler to die contact should be ok. If some part of the die still had no contact it should be dead by now.
And it seems to be only one side, the other one was still at 80°C, a huge difference of 14°C. What could cause this?

On taking it apart I inspected the cooler and cleaned it with distilled water. It had some screws in it, that seemed like they would hold a plate on top of the fin-stack and I unscrewed them but the plate didn´t move one tiny bit so I put them back and just left it at that.
Did I damage it by doing this? My guess now is that I loosened connection between some of the fins and one of the baseplates, so that the heat of one side can now only sink into a few fins if at all. Since I can not open the cooler I have no idea how to fix this. I might need a new cooler now.
One hint that indicates the heat source is connected to less mass then before is the quick rise and drop off of the temperature:

EDIT: btw I also have no clue why GPU-Z reads it on the Windows drivers, while it states clearly on the other tab that it uses nvidia 337.88 and SLI is working...

From 94 to 78 it dropped within 2 seconds.

This card really wants to bother me it seems.

 

[Mr.Scott]

It just shoots up faster then ever, especially on the one GPU that has the PCIe connector on the pcb. I aborted the test, it did not crash and since it does take a minute to get to 90°C I think the cooler to die contact should be ok. If some part of the die still had no contact it should be dead by now.
And it seems to be only one side, the other one was still at 80°C, a huge difference of 14°C. What could cause this?

That side will always be hotter. Less air flow and sandwiched between heat from the other card and it's own heat.

 

[Radical_Edward]

Wow that looks like some years worth of floof. Good thing you took it apart, did you put it back together yet?

I did, it's lived in filtered cases all its life, however it was in use until just this last fall in my wife's system. Guess it wasn't getting dusted out as much as it should have been.

All put back together and ready to go in my backup system once I get a mobo.

 

[Salty_sandwich]

I did, it's lived in filtered cases all its life, however it was in use until just this last fall in my wife's system. Guess it wasn't getting dusted out as much as it should have been.

All put back together and ready to go in my backup system once I get a mobo.

I have a graphics card given to me by an old friend an old Palit GF9800GTX+ / 512MB DDR3/256bit

who must have been quite a heavy smoker.
After i had taken card apart to remove years of fluff the card is nicotine stained ! i said to him crikey did you blow ya smoke into ya PC directly ?

everything, the motherboard, CPU heat sink, PSU the lot, covered in dark tar like crap lol

i see if i can find it (i think i still have it) and stick up a pic lol

 

[Dinnercore]

This thread is not dead, me neither I think.

I have trouble finding time for all my hobbies between life and its responsibilities. Now that semester break is starting once again I can get back to this.
The 9800GX2 is still alive, but the cooler seems definitly busted in some way. I´m looking for a replacement from a dead card and while I scan for a cheap score I will take a look at this guy that I found recently:

Dusty, a bit yellow and telling from the smell this card was owned by a smoker at some point.

7900 GTO from MSI. Rusty screws, 12 years old. Received it today, will test it next.

 

[Tatty_Two]

This thread is not dead, me neither I think.

I have trouble finding time for all my hobbies between life and its responsibilities. Now that semester break is starting once again I can get back to this.
The 9800GX2 is still alive, but the cooler seems definitly busted in some way. I´m looking for a replacement from a dead card and while I scan for a cheap score I will take a look at this guy that I found recently:

View attachment 115048

Dusty, a bit yellow and telling from the smell this card was owned by a smoker at some point.

View attachment 115049

7900 GTO from MSI. Rusty screws, 12 years old. Received it today, will test it next.

I used to have one of those...… awesome card!

 

[Dinnercore]

I wonder how the previous owner got it into his pc or if that bend slot cover was a result of shipping (top right on the bottom picture from before). It did not fit like that, had to bend it back in shape.
Once it was in it just came to live like it was at home in my setup. No long boot sequence or boot-loop like with the 295 or 9800GX2. It just works!

Latest driver for this card is the 309.08 for Win7 64bit.

Today we have 19°C ambient temperature. Idle power draw is sitting just around 100W, idle temp looking fine on 34°C.
One thing I noticed was how silent this card is compared to the other ones I had so far. If the case is closed it really does not stand out from any of the other case fans. I really like that!

But lets put some load on it...

And now i´m even more impressed. Given the age of this card and the slight 'patina' on the cooler I had expected worse. On top of that, the fan seems fixed speed and did not get any louder. I would have loved to have this one back in my old PC instead of the 8600GTS I had instead.

Load temps look perfect, 55°C is actually to good to be true. I´m suspecting that someone already took good care of this 7900 GTO before it got to me. Whoever that was, thank you but I´ll still take it apart. The pads look like the stock ones from what I can tell.
Well power draw is reasonable under load ~175W.

After it ran for a while I could tell by the smell this was definitly owned by a smoker but it is not unbearable for me. I will clean it anyway. If I can´t improve thermals I may atleast be able to make it shine again

 

[Mr.Scott]

Nice card.
Here are mine.

http://3dmark.com/3dm06/16869829

 

[E-Bear]

just dont suggest a midget be added.

Add a mini itx and it might be.

 

[Dinnercore]

Nice card.
Here are mine.

View attachment 115227View attachment 115228
http://3dmark.com/3dm06/16869829

Nice, I wish I had the box to mine... And seems like I need a 2nd one to get into SLI testing too

So many things I want to do, so little time...

Well back to my card, I took some time to build a new test system. I had plenty older hardware around to use and using my 1800x just as a testbench for 10 year old GPUs was a bit overkill. So I sold that system and bought an open benchtable. The rest of the money will fund future cards, thermal pads etc.. I have already used a full 15cm x 15cm sheet of 1mm thick thermal pad on these cards so far.

Some pics of the new system:

As you can see I kept the PSU, I really like this unit. That USB-link is very useful for me. Specs:

EVGA 780i board - Supports 3-way SLI
Intel E8500 Dual Core - Running @3.7GHz with stock cooler
Intel Stock CPU cooler
4x2GB OCZ DDR2 800
PSU Corsair RM850i

OS: Windows 7 64bit

Surprisingly the power draw figures come close to what they were on my Ryzen system, it´s just consuming ~5-10W more on idle and GPU load.
Of course the board got the same treatment as my cards, after confirming that it works I cleaned the heatsinks and replaced thermal paste and pads on chipset / VRM.

Now returning to the 7900GTO:

I pulled it apart and took a look underneath the heatsink.

Nothing special here, paste was old but looking very good still. Pads a bit yellow on the parts that had been exposed to smoke.

There is a little VRM heatsink that came loose by pulling out some pins with a spring:

The fan-cover and heatsink itself looked a bit nasty with brownish dust all over:

So I gave it all a good bath and cleaned it as best as I could. The pads and paste I wiped off the card:

Eww. Well it is what it is, cigarette smoke is not nice and the smell clings to everything it touches.

I got rid of all the material and removed the fan + cover from the heatsink too. Put it all in a bath of warm soap water. It looks better now and the smell is less but still not gone.
Before I forget to mention it, just by pure coincidence I found out that just by touching the card close to the PCIe connectors while it is running produced heavy flickering and weird distortion on screen. I took a look at the PCIe connectors of the card and cleaned them with contact-tunerspray, when wiping them off there was a lot of brown and yellow residue on the towel... That cigarette smoke stuff really does get everywhere.
Anyway, time for the more pleasant pictures:

The die had a really deep violet color that the camera did not pick up that well.
Right behind it on the backside of the card I found the usual SMD-city, compared to modern SMDs these are still fairly large:

I noticed something on the VRM components, there is an obvious height difference between the right one and the other two on the left. The original pads did not account for this height difference and just got compressed far enough. While that works I don´t find it an ideal solution.

Why not ideal? Take a look:

After wiping the old pads off you can see the stains where it made good contact with pressure and these stains are only where the two high components are. Might be due to higher heat from these two too, but I went ahead and tried to improve the situation by using 0.5mm pads and 1mm pads to even it out instead of hoping that pressure alone fixes it (as my pads are harder then the stock ones and don´t get squished that easy):

I did not get a good picture of it, but it makes a decently flat surface for the heatsink to sit on now.

Back together:

Oh those sexy heatpipes...

Back to testing I found some higher temps this time. Partly due to a very slightly higher ambient and the mainboard blowing warm air on the back of it but I´d say also due to the fact that the old paste was still in perfect condition and had a good spread. My paste probably has not cured yet too, I did not cycle it before the test because my time is limited.

Idle sitting slightly higher at 36°C with the ambient today at 20°C.

Under load I´m now at 63°C max. which is ~8°C higher then before. Again, paste might take some time to cure and spread. I´m still good with this result, 63°C leaves plenty headroom.
One thing I can do now is take shots of the cards in action on my benchtable:

I really enjoy having it up in front of me. Its not really quicker this way, since swapping a GPU in and out of case is no big deal either, but it feels so different to see it up close.
Oh and I have a place for all my case badges now

Lets see what card we will have next. Might be ATI-time again.

 

[Dinnercore]

Why can´t anything ever go the easy way in life. Well complaining doesn´t help, but I have some minor bad news.

My Camera broke. I was an idiot, tired and forgot that it was still sitting on the floor connected to the pc via USB cable, I fell over it and kicked it across the floor. It did not fall any height nor did it come to a sudden stop, it was just sliding across the floor. No visible damage but it no longer turns on. Battery symbol and SD-Card LED are the only two things lighting up when I turn it on, indicating that the battery would be empty but it is definitly charged. Tested another one, same problem.
I´m now in a really bad mood, I did not expect this thing to fail because of my stupidity, it still had so much lifetime left on the shutter and all.
This puts me in a spot where I can´t continue with my card collection, because my only replacement would be the cam on my Galaxy S4 and that thing is not suitable for quality pictures. I don´t have any money put aside just to replace my trusty DSLR, so I can´t just buy another one either... Gonna have to try and fix it or work double shifts for another two months... I´m so angry at myself. Why why why.

Well I got some last pictures out if it. Our next candidate for this program:

The mighty ATI X1950 XTX. Powerful DirectX 9 flagship card, most powerful single GPU for something like 3 months back in the day 2006. This time I even got the original box with it! It looks cleaner than I expected.

Found this funny sticker on the bag it came with, never seen that on one of my more recent GPUs as far as I remember:

Back from a time when most people were not as familiar with PC-hardware as they may be today and when additional power connectors for a GPU were still kinda new.

I already put it on my testbench and fiddled around with driver install, it refused to install the catalyst 10.2 Version from Guru3D for 64bit. I had to install some optional windows updates, they started to fail installing, after some restarts it figured itself out and then the windows driver did not want to go away. Even DDU could only remove it for a few seconds.
I then went to the AMD site and got the 10.2 version from there under the legacy section. I went into device manager and forced the windows driver to stop bothering me, installed the CCC and the driver for this card. Finally it worked. Card is running, but my head was so stuck with the dead camera on my desk that I forgot to take my screenshots.
Will take them later and add them like usual.

For now I need to get this thing back:

Another night with no sleep it seems -.- But I can´t live without a decent camera and all my equipment is for this exact model. Wish me luck.

 

[hat]

That sucks. I know all too well the feeling of breaking stuff due to my own clumsiness... though, nothing quite that expensive...

 

[phill]

What's happened to the camera??

 

[hat]

It's literally in the second line of his post...

 

[biffzinker]

What's happened to the camera??

Didn't like being kicked across the room? I'm sure @Dinnercore was kicking himself for the fatal mistake.

 

[Steevo]

If I had access to all the old machines I had built I could give you work for a year. Good on you, I remember my old cards and the work that went into them.

 

[king of swag187]

Bought a box of local parts, unfortunately most of them were bent to all hell/pins dented/capacitors missing, but I did get a working R7 260X, GTS 250, and a nice HD 5770

 

[phill]

It's literally in the second line of his post...

Many apologises, just looked at the pictures as was trying to do something else at the time.. Guy thing, tired and not looking at what I'm doing

 

[Dinnercore]

Many apologises, just looked at the pictures as was trying to do something else at the time.. Guy thing, tired and not looking at what I'm doing

No worries, as my camera can tell you I´m sometimes in the same state.

Bought a box of local parts, unfortunately most of them were bent to all hell/pins dented/capacitors missing, but I did get a working R7 260X, GTS 250, and a nice HD 5770

Argh dead hardware is never nice to unbox. Like the 8800s I got from someone who shipped them in a way that I could pour the broken-off SMDs out of the box
Those 3 working cards are a score tho! Grats on that, I would love to have a GTS250. They are not hard to find but somehow I did not yet come across an offer that would have got me interested. As a mid-low power card that thing seems to be still sought after for internet and office PC use.

If I had access to all the old machines I had built I could give you work for a year. Good on you, I remember my old cards and the work that went into them.

Oh I think I´ll find enough cards to keep myself busy for a year or more. There are so many models released and the era I´m focusing on atm is highly available on the second hand market. Seems like 10 years is roughly the timespan people stay attached to their old stuff and then they notice that it is just sitting in box somewhere catching dust and taking space and suddenly they want it gone.

Didn't like being kicked across the room? I'm sure @Dinnercore was kicking himself for the fatal mistake.

I´m still busy kicking myself mentally. I think I´ll not get over it for about a week.


News are tho, I took the camera apart up to the point where I would have to de-solder tiny wires to dig further. This is a step I´m not 100% confident I can reverse so instead I tried my best to find anything that could be broken but nothing was visible. I took all ribbon connectors (there are so many) out and put them back, cleaned the board and tried to get some life out of it. It did power on this time, display turned on too. But the joy only lasted so long as the display is broken and the camera is still stuck. I can enter the menu now but it does not detect lenses correct and still claims the voltage is to low. Due to low battery voltage it locks the release.
I think this camera is beyond my capabilities, but I got nothing to loose at this point so I will try to dig further into it. For now tho I need a replacement which means I might look into buying the same body used, I found some reasonable offers below 200$.

Sorry for off-topic, back to my lovely X1950 XTX. The idle values:

Once again I´m at 20°C for ambient, due to the low stock fan-speed it sits at 61°C in idle. Power draw is reasonable as well with ~117W, the card does run on lower clocks and voltages in 2D-mode.
And this power save state is not detecting Furmark as a full-speed 3D-application:

I do get my load numbers, but the card did not go full speed. I do want to stick with Furmark as the load test for now, so I will use these numbers but they are not accurate for these ATI cards. I might have to use some game or 3D-benchmark like 3D-Mark 03 or 05.
In Furmark however we see up to 80°C, shortly spiking over that until the fan-curve kicks in and drops the temp back down to ~76°C. And there we have one of the critic points for this card, the not very clever fan-curve has it ramping the fan up and down during load which causes some audible noise and tone difference. It´s not loud but the really noticable change every 20 seconds is annoying. I´d rather have it constantly high, as it is still just as loud as the stock intel cooler on my CPU and that mainboard fan.

To read the stock fan-curve I used ATITool, glad this thing exists. Afterburner does not support these cards at all.

Seems like ATI wanted to be as silent as possible, letting the card reach 80°C+ before even starting to bother the fan. But the fan speed is only checked every few seconds so it sits at 13% idle, load hits, it peaks in the mid 80s, fan is set to 70%, it instantly drops back down but the fan keeps going until it checks temp again after ~10 seconds. By that time the temp is already back at 75°C or less and the fan decides its time to idle again. Another 10 seconds of heating back up to 86°C and the loop starts again.

I was not happy with that, mostly with the high temps. I set my own curve to start ~5-7°C earlier on each step and launched Half Life 2 - Lost Coast. I ran the benchmark scene in full 1080p with settings on high, 4x AA and texture filtering max.. HDR on.

74FPS average is a really solid result for full HD, back in the day my monitor was like 1024 x 768.

With my own fan-curve the temp peaked at 79°C, fan speed hitting up to 56%.

And idle changed too, 6°C drop just by running it at 25% instead of 13%. And at that speed it is still not audible over my other fans. So I wonder why they choose to set it up like they did, I can´t imagine that 25% fan speed would hurt the life span in a significant manner.

 

[ST.o.CH]

Great old stuff here, sub for more,
Also thanks for share.