New love for old cards - [GPU restoration]

[king of swag187]

Would 10/10 not get a GTS 250, here they're a little less than the R5 240, which can go in basically any system with a PCIE slot and has all the newest AMD support, as well as Display port and half decent gaming perf. It's still a dandy old card, might stick it in a project me and a friend are doing at school, a old minecraft PC for as cheap as possible, but fun times. The 260X is in my brothers rig until I can figure out how to fit a Fury X in it, and the 5770 has a special place in my HP 8200, hanging out the side with a SATA to PCIE 6 pin. Streams stuff decently well, which I like and need.

 

[Dinnercore]

Would 10/10 not get a GTS 250, here they're a little less than the R5 240, which can go in basically any system with a PCIE slot and has all the newest AMD support, as well as Display port and half decent gaming perf. [...]

But I want one, I started collecting cards and I want to have atleast one of each, maybe two for SLI / Crossfire tests.


I could not fix my camera but got a replacement body. I can take photos again and spend some hours with the 1950XTX:



Initial look was perfect from the outside, pcb looked clean and the cooler performed like it should.

Upon taking it apart tho I finally found something to do for me, mainly the cooler itself had some dust build-up.
Take a look at that fan, 0,75 amps -> that is a 9W fan at full speed. Holy moly this can move some air. But I think the design of the fan as a kind of paddlewheel fights with air resistance more than any other design does too.



The pads look clean, they are still the stock ones and I´m going to re-use them. No need to waste stuff here, in contrast to the brittle white stuff that I find on nvidia cards, which have to be replaced most of the time due to them breaking into small pieces.
The whole cooler came off easy, the main unit with the fan is held on only by the metal X plate while the copper heatsink on those memory modules is held on with some screws.



The comb-like heatsink for the VRMs has two screws, but after taking them off it was still glued to the components, it felt really solid and I could spot some white pad type under it. I decided not to forcefully break it off and just leave it as is.

Thermal paste on the die was still mostly wet with a dry spot right in the middle. I can´t tell if this was replaced by someone in the past or if it is still the stock application.

On the back of it I found the self-destruct button along with the switch to re-boot the matrix. Not sure which is which tho.

The die and some additional photos:





The only dust I could find on the card itself was where air exits the I/O bracket.


Again I have to compliment the cooler for being designed in a proper way, as in keeping dust away from the pcb and collecting it, if at all, only on the heatsink parts. Well it could be that someone cleaned it before I got it, but he did a bad job on the heatsink then:



I no longer fool around with the small syringes of MX4, I got a big boy now. The last two 4g ones lasted for all the cards I have in this thread, lets see how far I can get with these 20g.



And its back together, still a bit wet but I will deal with that right now.



I DO NOT RECOMMEND THIS! I kinda sometimes think I should know what I´m doing so I dared to do this, but you definitly shouldn´t. Let cleaned parts dry before mounting them and plugging it back in!



I just let it warm up and evaporate the water trapped between the fins. Looked really cool to see the condensate building at the exhaust. And final shot of it back up and running:

Next up, the new temps.

EDIT: Here they are:

For the idle temp I waited until the card was dry and did my usual routine - heated it up just a bit (~75°C) with furmark and waited a while for the temp to settle back down. Same ambient of 20°C, this time we got just under 60°C. A small 1°C improvement that is confirmed by the 1°C drop on the pcb temp too. BUT still well within margin of error.

Load numbers did not change due to the fan curve managing the cards temperature, it wants to maintain something around 80°C.
This concludes the ATi X1950 XTX!

 

[Dinnercore]

Next card! Hello Ruby!

ATi / AMDs HD4870. The 512mb version. This will be likely the most recent series from ATI/AMD that I´d like to collect. It was competing with the Nvidia 200-Series and this model was reviewed as slightly ahead of a GTX 260 while much lower in price.

Again got it without the box, and shipped together with another card just loose in a simple box with some air packaging bags between them. The box was beaten and I was worried it might not have survived.

I can already spot more dust inside the fan and cooler as with the X1950 XTX before. The pcb has some minor damage on the corners opposite of the slot cover.



The back is looking clean and no damage. The whole board is bend just slightly, not sure if this comes from mounting pressure of the cooler or if it has to do with the scratched and beaten corners...



Doesn´t seem to bother it in any way tho, it boots fine and is working. This time using Catalyst 13.9, got it again from AMD. Upon starting the system the fan goes crazy at ~80% fanspeed and is really loud but moves a lot of air at the same time. It again spins up to 80% for a very short time when windows boots and the driver takes over.

In idle we can see what the two power connectors already suggested, serious idle power consumption around 160 W. The temps however hold where the driver or bios wants them to be with a low noise fan speed at 21%. I like how there are so many sensors that show up in GPU-Z, I´m a big fan of monitoring EVERYTHING. Be it useful or not.
Again ambient temp of 20°C, card is idling at ~66°C.

Load test shows that the card is still in good working condition. Core temp is fine at 75°C, the other temps we can see are ok-ish as far as I know about these things. VRMs do get hot and they are rated to tolerate that. The memory in the low 80s is still ok too, I´d say on some nvidia cards it runs the same or even hotter (the stuff on the back of GTX 295 single pcbs does get toasty quick) but I could never find a sensor read-out for those.
Healthy power draw peaking at 271 W. The fan is just audible at 33% speed but not noisy. The fact that it doesn´t need to spool-up the fan further shows me that the heatsink is still doing a good job.

 

[Dicfylac]

Is there any chance you may test one Radeon HD2900XT,
I had one back in 2007 until 2011, was my first flagship gpu.
I still have it but unfortunately it´s death.

 

[Dinnercore]

Is there any chance you may test one Radeon HD2900XT,
I had one back in 2007 until 2011, was my first flagship gpu.
I still have it but unfortunately it´s death.

There is a big chance I will get one and post it here.
Currently I don´t have one yet and I will first go through the cards I have ready to work on. But if I go hunting again I can keep an eye open for this exact model. I want to collect so many different cards and usually pick them at random so it doesn´t matter to me if I get the HD2900XT first or for example the FX6800.

 

[Kovoet]

I have a old 5870 lying around. You've given me an idea. I might just give this a go

 

[Dicfylac]

There is a big chance I will get one and post it here.
Currently I don´t have one yet and I will first go through the cards I have ready to work on. But if I go hunting again I can keep an eye open for this exact model. I want to collect so many different cards and usually pick them at random so it doesn´t matter to me if I get the HD2900XT first or for example the FX6800.

That would be great, nowadays is hard to get an old flagshish graphics card.

 

[DOM]

I have alot of old card sitting in storage ☹

 

[Mr.Scott]

I have alot of old card sitting in storage ☹

What, next to the boxes of baby clothes?

 

[Dinnercore]

I have alot of old card sitting in storage ☹

Then get them out and let them breathe electrons again! Well if you have a system to run them in... If you got watercooling in your current build it might not be that simple to just switch the new for the old for a day.
But I can tell you it may bring back a lot of memories and fun to run them again with your favorite game titles from the day. The familiar sound of those coolers alone does it for me...

 

[Dinnercore]

Please excuse the following short interruption of the main program:


I found a 9800GX2 for 15€! It was sold as dead and I can confirm that it is, but I was most interested in using the heatsink anyway. Now I can test my theory on the working 9800GX2 that I got and is still waiting for an update. The dead card here has some major issue with power delivery, I did not take it apart yet, BUT the lower board does not power up as in the LED on the back stays off, no picture from any display out, mainboard reports GPU error and just touching the cable resulted in the power indicator LED flickering from green to red. Will take a closer look at that later.


The 4870 got stripped and yep, much more dust in this one.



Enough to feed a small dust bunny family. This time I decided to test my 1.5mm thermal pads vs. these thick stock ones. I replaced all of the pads, VRM + memory with my Arctic Pad. 1.5mm on the memory, 1mm on the VRMs.
More screws this time and I don´t think the bend pcb came from mounting pressure, in fact without the cooler in place it was slightly worse.

Some of the stuff I pulled from the card:


Before I did anything else tho I took care of the beaten corners by dipping them with CA-glue.


The damage started to get worse, everytime it touched something on these damaged edges more fibers came out of the pcb. With this glue it is now sealed again. I choose CA-glue just because I had it around, I would guess any kind of glue that is not too aggressive would do. Like hot-glue for example.



^These are the things that manage power for the core while this:

Should be memory supply.

Does your card have VITEC?


Yo this new VITEC on my power stages kicks in at around 70 W and pushes the core to insane speeds, your boost V2.0 aint nothing compared to this! (sorry for the car-related joke)

I´m blue abedi abedei.




Back to cleaning the heatsink, there was a lot of floof inside too. Before and after:


And the dust-plagued fan, this time a 12W power blower. Getting close to hair-dryer level here.


I did not bath the heatsink unit itself this time and again just used a brush to carefully remove most of the dust from the fan. The fins felt really loose and got bend pretty easy, the heatsink on the X1950 XTX was much more solid in its construction.

Upon putting things back in place I had some trouble to align the baseplate with the copper-core for the die on the card. The problem is that the big heatsink on the die is completly seperated from the baseplate with the fan and has ~3-4mm room for movement. So when I put the card on the cooler and aligned the holes for the core, the baseplate holes were out of place and moving that plate into place meant potentially sliding the thermal pads out of place too!
Best thing you can do in that case is to lift it up again and check if pads are still good and carefully align EVERY screw hole before you drop the card on the cooler.



Now that it was proper clean and on new pads + paste I was curious to see if anything changed in thermals.

It does not look like it. Pretty much the same numbers as before, again within margin of error. Only thing I noticed was that the fan RPM is just a bit more consistent and rises less frequent. All the dust did not seem to bother it yet.

And my thermal pads work perfect as replacement for the stock stuff. That is good news.
Again the temp is decided by preset fan-curve from AMD, it wants to hold these temps and does so with ease at 33% fan speed. I did a little experiment with my own fan-curve this time using Afterburner.

This is what the idle temps looked like after I started my own fan-curve (letting it cool off from the Furmark test):

Just a single step higher in fan-speed, 27% instead of 22%, sees a whopping 15°C drop on the core and memory controller. Even the VRMs dropped 7°C across the board. When I´m using this card I will now set a custom fan-profile. Taking some °C off from 10 year old components seems worth the slightly higher noise. Which was still on par with what my other system fans produce.

BTW I have upgraded the testbench with a new cooler, the Asus V60! It does wonders compared against the intel stock one and fits perfect on this very size restrictive board.
On top of that I added one of my old HDDs to make room for some games and potential benchmarks. The small SSD was just enough for the system + basic things I need.

 

[Dinnercore]

The 9800GX2 is a big pain to work with. So many screws and the really sketchy ribbon cables that easily detach on assembly.

I have opened the one that I received as broken with the wonky 8-pin power but after taking a close look at it I can´t find anything wrong.


Someone did replace the thermal paste and decided to paste that SLI-ASIC too, which did seem to work more or less. The 8-pin itself has no weak solder connection as far as I can measure after probing it with my DMM. On both pcbs the VRMs resistances are fine, I see now visible damage. IF something is broken, it is again somewhere I can´t check right now. I might try and see if I can check any voltages on the side that does power up, but my guess right now is that one of the ribbon cables was loose.
Either way, the cooler is not the fix for the problem I have on my working card, it is again the thermal pad situation.

You see, the original 9800GX2 pads were really squishy and got compressed to a height of ~0,72mm. I only have 1mm or 0,5mm. The 0.5 does not make contact with the cooler while the 1mm is to strong to get squished enough by the weak mounting pressure that the tiny springs on those screws can provide...
So using 1mm pads I have contact between VRM, ASIC and memory modules -> heatsink BUT the paste on the core does not get squished enough, so the distance between die and heatsink is somewhere around 0.3-0.4mm which is too much for efficient heat transfer. This made the card/pcb soak with heat, as seen by the temps that slowly rise up to 70-75°C in idle.

To fix this I have to improvise, there are no 0,72mm pads on the consumer market as far as I can tell. I can either try to squish some pads by hand, or try to increase the mounting pressure by introducing washers on the screws. But I have to be careful not to give it to much and break something.

 

[silentbogo]

I guess I'll join the retro party )))
Found this puppy in one of the drawers in my office. Can't remember where I got it, but I'm guessing it ended up there cause someone thought it was broken...
Apparently all it needed was some dusting, pasting and lubig. Running like a champ with under 65C at full load. WD40 and no-name silicone lubricant worked some miracles on that fan (had to scrub the magnetic ring and the rotor from dust). Now it's clean like my cat's balls and it's nearly inaudible even at full speed!

Also dug up a reference Gainward GTX285, which is due for some cleaning tomorrow (and possibly vRAM replacement, cause as far as i remember it had some nasty non-GPU related artifacts).

 

[Jism]

Hacking the stock heatsink to fit the VRM's, kronaut straight onto the VRM's and a set of washers to tighten up. Result? Barely 40 degrees on 15 min furmark. When going 1450Mhz @ 1.175V and furmark "extreme" with Postfx enabled. the VRM's hit 60 degrees while the GPU power is over 260W.

I think alot of the one-heatsink-fits-all unnecessary heat up the VRM's and memory modules. In stock config this would easily rise to 65 degrees or so. Coud'nt think of doing PostFX turned on at all.

 

[Dinnercore]

I´ve finally done it! The 9800GX2 is ALIVE and KICKING. Have to celebrate that with some special music, straight from my heart. Dancing across my room atm. 4:52am and I don´t even care.



You have no idea how many hours went into this card and getting it to this point. I took this thing apart and reassembled it 5 times just today.

Let´s start where we left off, I initially opened it again to switch the heatsink and test if the one from the donor card was working better. Nope. After that I switched to the original cooler again, but tried to squish the 1mm pads a bit further, nope and on top it now did not boot and the mainboard threw GPU init. errors at me.
I took it apart again, checked if I broke anything or if the ribbon cable was loose, nope nothing to see. I put it back together with 0,5mm pads on one side and 1mm on the other to test the difference between the two and to see if 0.5mm might just be enough height to make contact with the heatsink. Oh and everytime new paste too. It booted and worked, no error and no trouble this time. I saw that the side with 0.5mm pads looked decent in temps, while the other one was ~10°C higher under load. So I took it apart again, went for 0.5mm on both sides, but it back together aaand...
Again it threw error, did not boot and one side appeared dead.

So, you guessed it, I took it apart again... (this process includes 37 tiny screws, 16 memory modules to cover, 2 chips to cover, VRMs and the ASIC) And saw nothing that would cause this. Ribbon cable was tight and in place. I then decided to put it on without the beaten cover on the outside, maybe this shortened something. And guess what, it worked fine without the cover on. Now I don´t know if that was really a short somewhere or some EMC problem, since the cover is connected to ground and they taped over the place where the ribbon bridge connectes the pcbs, possibly to prevent inducted noise. So to fix this and still save the original cover (all about the restoration here) I placed two non conductive thermal pads between that cover and each side and carefully adjusted the tape behind the ribbon connection so that it seats right above that.

Maybe it helped, maybe it didn´t but I have a working card! And finally new temps for my chart:



Oh how sweet these numbers are compared to the condition I received the card in last year. For idle I used the same fan speed fixed as last time, since it now idles with much lower speed. 53 / 54 °C, that is 5 / 7 °C lower than before, while the ambient today is just a touch lower at 20-21°C. But it idled fine before, lets throw Furmark at it and see if it starts a fire.



Before I aborted the test at 2:11 / 92 °C because temps kept climbing while the fan was maxed out on 100%. This time the temperature was the one that gave up and maxed out at 91°C on the hottest side. The fan was switching between 94% and 96%, so it had some room left. Given the card and Furmark being a power-virus test I´ll say this is decent.

Now I have a second card split in half on my table that has working VRMs and showed the same trouble like my card did with the wonky cover... Time to try and get a second working 9800GX2! But first some rest.

 

[Apocalypsee]

For the love of God please don't use Furmark. Everytime I see people using it I died a little inside, especially these old and delicate cards. Its nothing but a heat and power virus. Run uniengine or whatever, please, anything but Furmark

 

[Mr.Scott]

For the love of God please don't use Furmark. Everytime I see people using it I died a little inside, especially these old and delicate cards. Its nothing but a heat and power virus. Run uniengine or whatever, please, anything but Furmark

Thank you.

 

[Dinnercore]

For the love of God please don't use Furmark. Everytime I see people using it I died a little inside, especially these old and delicate cards. Its nothing but a heat and power virus. Run uniengine or whatever, please, anything but Furmark

I know that this is more pain then necessary and that I´ll never see this much heat in any real usage scenario but it does one thing very well: Testing the function of the cooler.
And it does a good job at being a reproducable load. With unigine benchmarks e.g. I can´t test these dual GPU cards because they don´t always work reliable due to SLI with stutters and load varying from testrun to testrun. This results in 2 different max. temps every time I run these.

For single GPUs this may be ok, but still is more difficult to use. Furmark is a dirty but quick and easy test. It is even used professionally by companies like Alternate (pc-system and parts dealer) for testing RMAs (EDIT: If I remember correctly some manufacturers even demanded this to be run first by us before they accept any returns). I used this test on cards like this 9800GX2 way back when I was working there during a school internship.

I can totally understand you all, and believe me I sweat more than all of you, after all I´m sitting here next to the cards screaming in agony for 10 minutes. But after these 20 minutes total (which is not a long period, like some people run this thing for hours) the cards will never have to do this again and I carefully use them only in games for a while, monitoring them so that they may last a little while longer. The ones I´m not using I store in special boxes with dehumidifiers that I control every month. Believe me I love my cards and I care for them.
But for once I will defend Furmark, if the cooler does not perform to spec or if there is a weak spot somewhere and the card is close to death it will most likely show up during Furmark. If a card dies like this it might not have lasted much longer either way.


EDIT: I made some progress with the 'dead' 9800GX2. It is up and running again, but only half of it.

Posted a separate thread for this and got the suggestion to try and flash the bios, and I´ll try that next. It does run like this tho, I ran the Half Life 2 - Lost Coast benchmark (found that title fitting for the situation ;P)



Solid 146 fps in full HD with some AA and texture filtering stuff. If half of it does turn out to be dead I can atleast try 9800GX2 3-way SLI

 

[phill]

For the love of God please don't use Furmark. Everytime I see people using it I died a little inside, especially these old and delicate cards. Its nothing but a heat and power virus. Run uniengine or whatever, please, anything but Furmark

Thank you.

I prefer Heaven or even Catzilla.... Just something different to watch other than 3D Mark.. That said 06 is pretty ok, same as Vantage I guess....

 

[Jism]

For the love of God please don't use Furmark. Everytime I see people using it I died a little inside, especially these old and delicate cards. Its nothing but a heat and power virus. Run uniengine or whatever, please, anything but Furmark

It's a stress test tool, just as AtiTool was once before. It's a very quick and effective way of testing clocks. voltages, cooling and all that. Simular as running Intel burn test on a OC'ed CPU just to test the worse possible scenario. I'd prefer doing this and any decent designed card should be able to handle it very well.

My RX580 did a whopping 260W core usage and the VRM's where pulling as good as 20 amps over one single 8 pin connector. That's the testing what i wanna see when going for a max 24/7 OC. When the cooling is properly and the supply of power is sufficient, nothing much that could go wrong. Just dont select POST-FX because THATS the power virus you are saying. In normal furmark we're talking 140 to 180W of load.

 

[Dinnercore]

It's a stress test tool, just as AtiTool was once before. It's a very quick and effective way of testing clocks. voltages, cooling and all that. Simular as running Intel burn test on a OC'ed CPU just to test the worse possible scenario. I'd prefer doing this and any decent designed card should be able to handle it very well.

My RX580 did a whopping 260W core usage and the VRM's where pulling as good as 20 amps over one single 8 pin connector. That's the testing what i wanna see when going for a max 24/7 OC. When the cooling is properly and the supply of power is sufficient, nothing much that could go wrong. Just dont select POST-FX because THATS the power virus you are saying. In normal furmark we're talking 140 to 180W of load.

Ok I kind of agree with you but have to say for OC stability this test is not good. Furmark runs stable with clocks that would crash ingame. Just like you can run insane speeds on the GPU-Pi benchmark that would crash anywhere else. It is good for testing the cooling and if the cooler is up to the task under all circumstances. Or to simulate the additional heat on the components that you might see in higher ambients during summer when its currently winter.

But like the others said, please don´t use this as 24/7 stability tester. To test OC for stability run a short Furmark run under full supervision to see if the temps get critical and then run your 3D-Benchmarks and games to test.
Don´t run Furmark hours on hours or even overnight while you sleep. That kills VRMs by pushing them slightly out of spec and reducing the lifespan significantly if you aren´t there to stop the test when something gets too hot.

 

[Jism]

A properly designed card woud'nt run out of spec, that is the whole point. They have a OCP and proberly a bunch of SMD based fuses on the card itself. The core is actually shutting down at a certain threshold as well. I'd agree with you that furmark is'nt the tool to be left alone 8 hours or more but running for bare 5 minutes aint going to cut it as well.

I live in portugal, and i had a terrible hot summer with 46.5 degrees outside lol. When having an office of over 35 degrees ambient it's running into it's 90's easily. So yeah you want to have a 24/7 sustained OC that would defenitly pass in any circumstanes. I cant have a crash and the driver resetting the GPU in the middle of a game if you know what i mean.

 

[Dinnercore]

As some may have read already, the EVGA 9800GX2 is gone for good now. Last thing it did was playing some Half Life 2, I´d say that is a nice way to go. One day I might want to go like that too...

Now it may be apparent that I´m into dual GPU (VPU in this case) cards. Which is why I had to take a look at ATi / AMDs take on mGPU. Apologies to @ST.o.CH , the HD2900XT will have to wait until I come across one. You are right, they are not very common to find these days. Even less than the 1950 XTX.

I found this next card under the 'for free / as gift' section. I was a bit too far away to drive over and pick it up, so I made a deal with the owner for shipping.

It came with the original box and all the stuff in it. You can instantly spot that the only original part that is missing is the cooler. Instead this card was already fitted with an aftermarket solution from Zalman. Two Zalman VGA coolers for the dies, and on the original plate that is covering the memory modules some extra little heatsinks in blue. Suits the card quite nice. All the other heatsinks, like for the VRMs, are stock.

I´ll see when I find time to take care of it, currently using my benchtable to test out another mainboard and some CPUs.

 

[Dicfylac]

Now it may be apparent that I´m into dual GPU (VPU in this case) cards. Which is why I had to take a look at ATi / AMDs take on mGPU. Apologies to @ST.o.CH , the HD2900XT will have to wait until I come across one. You are right, they are not very common to find these days. Even less than the 1950 XTX.

No need for apologies, few time ago I had look for an HD2900XTX and didn´t find any, but came across one GTX 9800 GX2 for 40€, one HD 4870 X2 for 35€ and one GTX 295 simple pcb which price can´t recall, didn´t took any because wasn´t interested.
BTW that HD 4850 X2 is very nice .

 

[Dinnercore]

Between mainboard swaps I found some time to test the 4850X2:



The fans plug into the mainboard and as such are controlled by it. This means I have to set them to a fixed speed and can not make any load adjustments, so I set them to 80% all the time.


33°C idle temp @ 20°C ambient. High fan-speed plus aftermarket coolers = low temps. I have to add 20W to the power draw figure, because this board was DDR3 instead of the DDR2 before and idle draw was lower. Same CPU.



For this load test I used the GPU-Z render in fullscreen to generate some heat. I wanted to try the suggestion of running a game instead of Furmark, but had trouble to get the Half-Life 2 benchmark to run in Crossfire. It only used one GPU even tho Crossfire was activated in Catalyst software.

Again decent temps, 48°C max on one core and highest overall sensor at 53.5 °C. Power draw with adjustment = 294 W

Will take it apart next.