New love for old cards - [GPU restoration]

[Dinnercore]

Hello dear people out there!

I´m Dinnercore, on this board for some days now and enjoying myself around here. I thought of opening this thread to celebrate my new found love for 'older' GPU´s.
They once had been the latest tech and were in use for a long time, decades in some cases. Now they land on my doorstep, beaten, dusty, still trying all they have left in them but in the tech world time is moving fast. Consuming to much power, while not suitable for any modern game they desperatly look for a purpose and I want to give them one in my machines.

But before they can live in a happy retirement under my watch they need some serious attention. I put a bit more time and effort into them then just spraying air and throwing new paste on. I want to restore them as best as I can. I find the work very relaxing. And I can relive all the memories that come up with the dream hardware from my childhood.
On top of that, I want to try different aftermarket coolers, watercooling etc.. Multi-GPU setups and some OC adventures.

What´s in it for you? A closer look at a lot of pcbs and coolers from different times, the comparison of before/after if you´re into that. Every now and then I plan to show them running in a system (not necessarily a perfect time-matching system for now, I lack the time and space for that much hardware).
The rough time period you can expect is anything from the late 90s up to 2010.

NOTE: If you like any of the photos I took and want a high res version, you can ask me and I´ll see where I can upload it for you. I just can´t upload all photos in original 4K res because of filesizes.
Same goes if you want to know or see a specific detail of one of my cards.

The following table contains my collection and results so far:
Please note that temps are in delta T over ambient due to fluctuation in ambient temp. Power draw is measured by my PSUs 'Power-in' reading.

GPU​	Cooler Type	Idle before	Load test before	Idle after	Load test after	Temp improvement	Power draw Idle	Power draw Load	Stock Core	Stock Memory
GeForce GTX295 Zotac 1792MB sPCB	Stock Cooler	20 °C	n/a	17 °C	48 °C	-3 °C	130 W	400 W	576 MHz	1008 MHz
GeForce 9800 GX2 Leadtek 1024MB	Stock Cooler	39 °C	OTP	33 °C	70 °C	-6 °C	175 W	370 W	602 MHz	999 MHz
GeForce 9800 GTX+ XFX Black Edition 512MB	Arctic Accelero Xtreme 9800	12.9 °C	n/a	11 °C	36 °C	-1.9 °C	125 W	275 W	785 MHz	1150 MHz
GeForce 9800 GTX+ EVGA 512MB	Stock Cooler	12 °C	55 °C	11.7 °C	51.7 °C	-3.3 °C L/-0.3 °C i	98 W	255 W	738 MHz	1100 MHz
GeForce 7900 GTO MSI 512MB	Stock Cooler	15 °C	36 °C	16 °C	43 °C	+7 °C L/+1 °C i	98 W	176 W	650 MHz	660 MHz
ATi Radeon HD4870 512MB	Stock Cooler	46 °C	56 °C	46 °C	56 °C	0 °C L / i	161 W	271 W	750 MHz	900 MHz
ATi Radeon HD4850 X2 1024MB	2x Zalman Quiet VGA Cooler	13 °C	28°C	tbd	tbd	tbd	172 W	294 W	625 MHz	993 MHz
ATi Radeon X1950 XTX 512MB	Stock Cooler	41.4 °C	61 °C	39.9 °C	61 °C	-1.4 °C i	117 W	185 W	648 MHz	999 MHz
ATi All-In-Wonder X1900 256MB	Stock Cooler	31.3 °C	59.1 °C	31.6 °C	56.9 °C	-2.2 °C L/+0.3 °C i	125 W	175 W	500 MHz	477 MHz

Additional information for this table: n/a usually means that I could not take this measurement because of e.g. the load test crashing or cooler parts missing. The Furmark load test will always be 10 minutes on aircoolers, since they don´t take that much time to saturate. Note that for some more delicate dual GPU-Cards I will try to avoid Furmark in the future. On future watercooling I plan to use a Mo-Ra3 loop just for the GPU/s. I will make some tests as to when the water temp has settled and plan a time accordingly.
Ambient for calculating delta T is measured right at the front intake fan in the case.
All my repasting is done with Arctic MX-4.

Anyway, let´s start this off with my latest puppy:

An XFX 9800 GTX+ Black Edition.

Already came with an Arctic Accelero Xtreme but the voltage regulator cooling must have fallen off. There is only the glue residue left. Oh well I´ll see what I can do.

Interesting mod (not by me) on the cooler for the GPU-status LED cable.

This is nasty, but could be worse.

Atleast the thermal pad could tell me what memory this card has:

The card is tested and runs fine, gets picked up by driver and GPU-z and it can run 3D-load however I did not make an endurance test and I´m very happy with that decision now that I see that VRM cooling is no longer present. It may be fine like this, but I don´t want to risk it.
Next post from me will be the result, after I got some thermal glue.

 

[dorsetknob]

Keep them coming

 

[phill]

Subbed

 

[Mr.Scott]

Subbed also.

 

[DeathtoGnomes]

Subbed too!

 

[rtwjunkie]

I’m in! Almost like pr0n.

 

[DeathtoGnomes]

I’m in! Almost like pr0n.

just dont suggest a midget be added.

 

[Urlyin]

take some pics when their all cleaned up afterwards... perhaps some GPU temps before and after? Welcome to TPU Dinnercore

 

[Dinnercore]

Oh wow, thanks! Wouldn´t have thought so many would like to see these cards.
The suggestion with temps is nice, I´ll see how I work that into the whole thing. Maybe I could create a spreadsheet in the first post or sth.. But I´m worried about the fluctuations in data by my ambient temp. It´s all over the place in the next weeks, like today it´s 26°C in here and if I open up a window tomorrow it´ll be more like 20°C.

Anyway, some progress on the XFX card:
I´m not happy with those stains from the paste that someone left there (e.g. in the middle of the 4 top left memory modules). It is barely removable with alcohol.

EDIT: these noisy smartphone cam pictures dont do it justice, from now on I´ll bust out the DSLR no matter what. (replaced old pics in this post already)

This part is looking a little bit better now.

And do you prefer the dark side....

Or the light side?

And with my CSI super zoom I tried to catch some strange marking on the die:

There is two of those squares, they seem perfectly aligned on the same height. Was this where the tool grabbed it after it was cut from the wafer? I have no idea.

Full assembly will be tomorrow when I get the glue for VRM cooling. I have sourced some fitting heatsinks from broken plasma TV pcb´s for free, but forgot that I will actually need something to hold them in place...
I´ve got 3M double sided thermal tape, some thermal glue and some xbox branded thermal glue-pad thingy incoming, I want to test which of those things work better.

And a teaser for what´s coming up next:

 

[Urlyin]

not sure if you're aware but I found acetone to work better than alcohol... pics look fine.

 

[qubit]

Hey, I look forward to your restorations, dinnercore. This is properly nerdy so the qubit approvez.

 

[Dinnercore]

Time for the next update. As life is, I got some good news and some bad results with one of my thermal glue solutions.

And thanks @Urlyin for the hint on acetone, did not try that yet and will give it a go in the future.


Cleaned cooler.

Next up, my try on the VRM heatsinks. I tried the first and cheapest solution to stick them on first: A double sided 'thermal conductive tape' from some chinese no name reseller. 10€ for a 25m roll is a bit expensive for just tape, but much to cheap to be anything that can transfer heat. It is however really really thin so I hoped that it would be ok. It isn´t. I interrupted the long term load test after 5 minutes, because the backside pcb was burning to the touch, while the heatsinks I could reach felt cold or lukewarm at best.
Lesson learned. I now have to take it apart again and try the second option from a company that does stuff for modding consoles. It´s a 100x100mm sheet, only 0.15mm thick and rated at ~3W/mK... The instructions on it suggest that the glue on it is a bit tricky to work with, I hope it will do. The 2 compound glue I ordered takes 2-3 weeks shipping.

Fully assembled, and by the time im typing this already on it´s back and loosing screws again

Temps!:
Ambient 24°C / Idle temperature now 35°C. Before taking it apart the idle was at 38-39°C with an ambient temp of 25,6°C.
I did not win much there, but it´s the feeling that counts.

And right after I cut short the first endurance test due to slight worry about VRM:
(Looks like it would have settled at 49°C anyway).
Full load test with furmark has to wait until I get those damn heatsinks to do their job. No idea if the card can run without, the reference cooler for this card seems to be lacking there but every other 9800 I saw online had some.

Another thing I want to monitor for this one and all the following cards is power draw. They will, if possible, all run on the same machine with the same config. I will measure power draw at the wall in idle and under furmark.
This one is for now: 124W @ idle // 234W @ GPU-z render thingy.

Finally, look what the previous owner send me without a word:

Aww yiss, DHE-stock cooler. I love those. They are my absolute favorite for looks. Especially the ones from around 2008-2010 with rendered artwork on them.

That´s it for today, I´m going back to testing these xbox thermal sticky pads.

 

[theFOoL]

That Light-Side photo should be in a Comp-Por¦ Magazine

 

[Mr.Scott]

For your heatsinks, no tape.
Your favorite non conductive TIM in the middle, and a dot of superglue on opposite corners.
Works titties.

 

[Dinnercore]

Final numbers for the XFX 9800 GTX+ Black Edition (@21°C ambient):

10min Furmark: 57°C // 275W Power in
Idle after Furmark: 32°C // 125W Power in

And @Mr.Scott that is another great idea. I was a bit paranoid to use superglue, I always have 100 what-if´s in mind (superglue and the stuff inside the coils fighting each other, glue to solid for temperature cycles = cracking and falling off etc.) But if it works that´s great news.
For now tho the adhesive worked.

A question for anyone who knows how to do things properly in a forum, what would be the best way to implement a chart for temps and power draw in this thread? I can´t edit my first post, so no adding there. I could just make another post, but will loose the ability to edit that one too. I could create one somwhere and link to it, but who would find that link between all the other posts?

PS: If anyone wants to see a specific test or benchmark, just ask and I´ll see what I can do.

 

[hat]

Not sure why you can't edit your post. Under your own posts, there should be an Edit button, near your signature (if you had one) and whre the Report button is for other's posts.

@FordGT90Concept posts tables more than anyone else I know of on the forums

 

[rtwjunkie]

A question for anyone who knows how to do things properly in a forum, what would be the best way to implement a chart for temps and power draw in this thread? I can´t edit my first post, so no adding there. I could just make another post, but will loose the ability to edit that one too. I could create one somwhere and link to it, but who would find that link between all the other posts?

There is a time limit on editing. PM a Supermod and see if they can extend your editing period on your posts in your own thread. Or it may be on an as needed basis. It's also possible I may recall that W1zzard keeps that locked down for himself. In any case, he doesn't usually have a problem if you explain the need.

 

[Jetster]

Had this for years. Never been taken apart. Should I?
I tested it a couple of years ago. Its only scores a 400 on Heaven 4.0. My GTX150 ti scores an 800

Someone want to get in on the shoot

 

[hat]

Heh, I've got a BFG GTX 260 Maxcore 55. It served me well for a while, then I got a 5870 and tried to sell it, but nobody bought it, so in the parts bin it sits.

 

[FordGT90Concept]

The first ATI Radeon (yes, that's a picture of my card):
https://www.techpowerup.com/gpudb/3024/radeon-sdr-pci

Doesn't need to be restored because of how little it has been used.

 

[biffzinker]

Heh, I've got a BFG GTX 260 Maxcore 55.

Still have my BFG Geforce GTS 250 (later version of the 8800 GTX after the 9800 GTX)
(but in a roundabout way I lend it to my brother for a Xeon build (Ivy Bridge) I donated to help him out on moving from his older Athlon 5200+ build.)

 

[Fouquin]

Still have my BFG Geforce GTS 250 (later version of the 8800 GTX after the 9800 GTX)

Eh, close. GTS 250 (G92B) was a rebadged 9800 GTX+ with slightly more reserved clocks and double the VRAM. G80 gave its last with the 8800 Ultra and GTS Core 112, G92 lived on for a couple generations beyond.

 

[Dinnercore]

Time for the next card. And chart is otw I hope, I contacted a mod via pm.

This card is giving me a real headache:
Zotac GTX 295, a dual GPU card.

It looks fine on the outside. But already gave me trouble just getting it to boot.

This one will be a longer story for sure...

Got it running into idle, but have trouble getting it to work under load in SLI.
Idle Temp @ 22°C ambient: 41°C / 42°C
Power draw: ~130W

For those interested in my troubleshooting journey with it (and it´s not over yet):

My testsystem for reference:
Ryzen 7 1800x
MSI X370 Gaming Pro Carbon (I hate this thing, never buy cheap)
16GB (2x8) DDR4 G.Skill Flare X @ 3200 14-14-14-34
Corsair RMI 850W PSU
120GB SSD

Got the card like usual, stupid me did NOT use DDU before changing the card from the GTX 9800+ to this, thinking oh it´s the same driver version it may be fine for now...
Well before it even got to the driver stage it didn´t post. No signal out. PC stuck, every 3rd power on it initialized keyboard and mouse, but still no signal. Fan spinning 100% on the GPU, green LED on.
Alrighty, off to a great start. Next thing I did, instead of using 2 seperate PSU cables, I used a single Y for 6+8pin.
Oh wow look at that it posts and boots. Wait a minute, why does my memory initialisation now fail?! Mainboard boot looped 3 times, went into standard config. Ok. Deep breath. This will be a long day.

Now it did boot. I got into OS (Win7 64Bit), checked that the card was picked up by driver and GPU-z. Went just fine. Scratching my head about why my system can´t hold OC settings or my ram profile now. So I rebooted, went into Bios, tried some higher voltages all around. Nothing, as soon as PC starts it completly shuts down, goes back up 3x (Mainboard retry setting) and then just proceeds with stock settings. Alright then. Time for a BIOS-Update.

BIOS-Update done, oh my god it finally starts without any delay at startup like it did before all the time. Wow thanks MSI, I can finally set subtimings! Just about a year late. Went into windows, everything still ok? Nope. BSOD within 2 minutes / related to me removing USB-Stick. Blaming the pci.sys as the fault. Well i´m not done anyway. Back into BIOS, getting all my settings back in order after the update, then I updated my Chipset-Drivers too. Time for some coffee.
Finally it seemed stable. Took the Idle temp measurement and went ahead into Furmark. Single GPU runs fine. SLI activated, 20 seconds in @64°C it died again with a BSOD. This time nvlddmkm.sys is stated as the cause.
Since it crashed pretty much related to the load I hope it´s not the card that reached EOL. The error code hints more towards driver tho, I hope I can fix this with DDU and a clean reinstall after all the changes I made. However, due to the possibilty of it being a heat/stress issue on the card in it´s current state, I´ll now go ahead and take a closer look at it before I do more testing.

Had this for years. Never been taken apart. Should I?

It doesn´t look bad from the outside, but since I see a cat around it there could be a lot of hairs inside. Not the nvidia hairworks you would want.
Maybe plug it in, check if it works and if temperatures seem reasonable. Does the fan spin, what noise does it make etc..

I have great memories of my GTX260, had to rma the first one I got due to cooler not making good contact. It was the first upgrade for my first pc!
And I need a GTS250, I would like to have the whole 200 series family. Got a 260 and 295 now.

 

[Mr.Scott]

Driver issue. 295 was always driver picky.
332.21 I know works.
No longer have a 295. Just a couple 260's (65nm and 55nm), 275FTW, and 280.
Really want to find another.

 

[Tatty_Two]

Op..... I have unlocked your first post so you can add/edit there should you wish.