14th subject: Nvidia Quadro4 980 XGL. One example of a high-end card not using tantalums, and being worse off for it. Three presumably polymers, 150 uF 16 V, and a bunch of wet surface-mount alus. Some have 12 volts across them.

One of the wet alus had obviously leaked, and there was a fishy smell elsewhere as well.

All 6 V wets replaced with 10 V surface-mount tantalums. A 470 uF 16 V and a 1000 uF 10 V wet replaced with through-hole wets, Nichicon UPW and Panasonic FR.

Works. The presumed polymers need replacing at some point, maybe with low ESR tantalums.

15th subject: PC Chips M577. Various wet through-hole alus, 10 uF, 47 uF, 100 uF, 470 uF, 1000 uF, 1500 uF. G-LUXON. A random selection measured ok for capacitance and ESR.

All <= 47 uF replaced with through-hole 25 V tantalums, others with Panasonic and Nichicon wets.

Works. Some molten plastic on the AGP slot.

Tantalums harvested from a 486 laptop. Not all of them, but most.

11 uF:   2
22.2 uF: 2
34.7 uF: 4
46.8 uF: 1
510 uF:  19
622 uF:  1
747 uF:  5
8100 uF: 1

All measure within 10% of spec in capacitance, ESR in the 0.5-2.5 range.

16th subject: ATI Radeon 9600. A history of multiple leaked surface-mount wet alus, covering the board in residue and artefacting the output. All removed and replaced with surface-mount tantalums. The four Nover 470 uF 16 V through-hole wets were replaced with Nichicon UPW.

Works.

17th subject: Matrox G550. Ties with the G400 MAX as the best 1990s gaming card from Matrox, though some would say the better of the two. Ten 10 uF 16 V surface-mount wet caps, all had less than five volts across them. Also some tantalums.

Wets replaced with 20 V surface-mount tantalums.

Not tested but probably works.

vvbee wrote on 2025-04-26, 08:04:

16th subject: ATI Radeon 9600. A history of multiple leaked surface-mount wet alus, covering the board in residue and artefacting the output. All removed and replaced with surface-mount tantalums. The four Nover 470 uF 16 V through-hole wets were replaced with Nichicon UPW.
Works.

The artifacting is almost always a result of a failed GPU chip on these... and that's a direct result of the card being passively cooled and running too hot.
My advice for anyone who has these Radeon 9550/9600 cards with passive coolers: either install a small 40 mm fan directly onto the heatsink or have a large fan blow air on the side of the card. The fans don't have to be blowing at full blast / 12V / max RPM... just move a bit of air across the card. This will dramatically lower the GPU heatsink temperatures. I usually go with the side fan option, as that also tends to cool the RAM too.

As for leaky caps on these cards - have not seen a single one with such. Not faulting you for replacing those Nover caps, though - they are a pretty crappy brand.
Nichicon PW is a little too conservative. The ATI-built cards use Nichicon HC, which is a step-up from the HD series... and HD is a step-up from the HE series, which themselves are a step-up from the PW series. So a better suited series from Nichicon at this point in time (since HC series are no longer made) would be the HW series. HE and HD will do too, of course. PW is more for older half-bridge PSUs (Fortron / Sparkle / SPI / FSP users take note 😉 ).

vvbee wrote on 2025-04-24, 05:40:

14th subject: Nvidia Quadro4 980 XGL. One example of a high-end card not using tantalums, and being worse off for it.

Like with the Radeon 9600 above, I'd say be more worried about the cooling on these cards than the caps. In particular, that small stock heatsink is one reason these cards die. The RAM chips also run pretty hot and should actually use some heatsinks... on both sides of the card.
And again, I have not seen a 980 XGL or a TI4400/4600/4800 with leaky caps - at least not from nVidia itself. IIRC, EVGA used some rather spotty SMT electrolytics that had pretty lousy ESR specs. But the worst was WinFast / Leadtek with some crappy GSC/Evercon caps used for the GPU and RAM output rails. On the positive side, at least they put a nice big cooler and RAM sinks on the card... and probably the reason why my Winfast TI4400 still works without a hitch, but the various stock-cooled TI4400's and 4600's I have are artifacting.

As for the SMT polymers - those are good and probably won't need replacement for the lifetime of the GPU.

Anyways, good work with the rest of the cards / boards.

I myself plan on treating an eVGA Riva TNT2 in a similar way to your cards, but with SMD MLCC's (multi-layer ceramic caps). Basically the card has a lot of small 10 uF 16/25V through-hole electrolytics that are getting in the way of fitting a larger passive cooler. The stock heatsink is a joke and has a super-tiny fan that's extra loud (it's running on full 12V, so yeah.) It's not a card I care that much about (or at all, actually)... but I use it a lot for testing older motherboards with AGP (universal AGP and older). It's particularly annoying after several hours of a board running through stress tests - that tiny cooler on the GPU drives me nuts!

For that Radeon 9600 multiple failed capacitors looks to have been the issue, though I don't know why they failed. Improved with one or two replacement surface-mount cans, somewhat blurrier 2D and occasional geometry errors in 3D with some vintage ELNAs and Panasonics, and works without issue with tantalums. But of course you never know. Their TDP is pretty low though and this particular one doesn't get very warm on the test bench with side airflow. It could be a Mobility version, can't remember.

The 980 XGL is a PNY, the capacitors didn't look very high quality anyway. The back gets fairly warm, it could use a back spreader like some of the GeForces have. Same with the TNT2.

vvbee wrote on 2025-04-30, 00:20:

Their TDP is pretty low though and this particular one doesn't get very warm on the test bench with side airflow. It could be a Mobility version, can't remember.

Yeah, as long as you have a fan blowing any amount of air over the card, it will run decently cool. That's what I have on mine - an 80 mm fan connected to 7V (for reduced speed, so it's not very loud), which doesn't move that much air... but even that is enough. If I disconnect it, the card's heatsink does get hot enough to almost burn skin. So these passive cards definitely need a fan nearby to move air over them.

vvbee wrote on 2025-04-30, 00:20:

The 980 XGL is a PNY, the capacitors didn't look very high quality anyway. The back gets fairly warm, it could use a back spreader like some of the GeForces have. Same with the TNT2.

Yup, these cards have oldschool non-flip-chip GPUs, so they also tend to cool a lot through their BGA solder pads on the bottom (and thus the PCB.) On my WinFast TI4400, besides the big heatsink on top, there is also a large plate on the back with a thermal pad right behind the GPU BGA - the one time WinFast got the cooling solution right. Now I also have a 8600 GT from them and that one is completely the opposite... but I'll leave that for another thread. 😀

Useful to also measure the temperature. Hot enough to start burning skin could be as low as 45 C or so.

vvbee wrote on 2025-05-01, 08:32:

Useful to also measure the temperature. Hot enough to start burning skin could be as low as 45 C or so.

About 55-60C for me. Checked with thermocouple. 😀
The thing is, if the heatsink is that hot, you can bet the silicon underneath will be about 5-10C hotter.
Just my observation from (obsessive) heatsink modding.

Great thread! I won't restate what I've said in the past in great detail, but having put power through hundreds of devices with many thousands of tantalum capacitors, I have only had three failures (one exploded, one sparked, one was simply dead short and stopped the system from turning on) in the 30 years I've used PCs, with the past 10 being heavily focused on retro PCs.

I will always be slightly leery of powering them on the first time because of the potential for a spectacular failure, so when I'm testing 30+ year old cards that likely haven't had power through them since the 90s, I tend to put something between me and the caps just to be safe. Once they have been tested for a few minutes though, I basically consider the card as good as new with no need for a recap for the foreseeable future. Unlike cards with aluminum eletrolytics... with those I'm always suspicious that any tiny graphical\audio anomaly I experience could mean that all the aluminum caps need to be replaced. In reality, even they rarely need replaced either (the problems often lie elsewhere), but overall it is far far more likely that they do, vs tantalums.

I have harvested a lot of tantalums (from devices headed to scrap) over the years... though admittedly I haven't done this in a while. I should probably harvest some more since I've got a pretty big pile of junk cards\boards accumulated right now. With a heat gun for the SMD caps and a Hakko FR-301 for the through-hole caps, it is a breeze! 😀

I'm seeing 55 C on the heatsink for the Radeon 9600 after five minutes of idling with zero forced airflow on an open bench. 65 C after a loop of the game tests in 3DMark 2000, and 67 C after idling for five more minutes on the desktop. Falls to 50 C with light manual fanning in the general area and to 40 C with heavy manual fanning. Ambient around 23. Not that I'd want my retro gear sitting at 70 C but not too bad for passive convection under load, also shallow temperature gradients and next to no thermal cycling so shouldn't be a problem for BGA cracking. Not great for the capacitors though.