Blast from the past #2: MSI 694D-Pro A ver:1.0

[Aetherone]

With the success of my Asus A8N32-SLI Deluxe recap project, I'm turning my attention to the other benchmark board in my history - the venerable MSI 694D-Pro A.

Wayyyyyy back in the day, I paired one of these (in protoype orange PCB no less) with a pair of 366 Celerons for a world of fun. That board was flawless from day #1 (well, aside from the usual VIA issues with sound cards and Promise getting narky about additional HDD controllers). Wish I'd never sold it off.

The current board is one of a hundred examples from a teaching college near the IT shop I used to work in. All of them died from bad-cap plague.
This board was hung onto for years as a trophy/example as to why we were preferring high-end boards with solid caps. Its been non-esd-safe handled by hundreds (if not thousands) of customers and yet, with a bodgy recapping, sort-of mostly worked again for a few years almost a decade ago.

Thanks to my bodgy recapping, I'm not sure exactly what used to be on this board - given the handling history, I had no idea if the board was working so I pillaged caps from all over the joint to repair it. The conclusion then: CPU's, PCI+AGP, memory all seem fine - weeks of uptime. Promise controller must be disabled or the system is crash happy (ESD damage maybe?).

This time around I'd like to do the job properly - with correctly sized and lead spaced caps that aren't wildly over-specced.

Looking about the forums, there's a few threads that are useful
Bushytails 2004 "MSI 694D Pro V1.0 recap"
Bushytails 2005 "MSI 694d Pro-AR recap" (lots of lovely helpful photos, thanks!)
Bushytails 2005 "MSI 694d Pro-AR recap #2" (again lots of lovely helpful photos, thanks!)

Working backwards with these for reference, I've sorted
6 Tayeh 2700uf 6.3v - replaced today with Nichicon 3300µF, 6.3 V, HM Series
8 Chhsi 1000uf 6.3v - replaced today with Rubycon 1000µF 6.3v ZLJ
3 Chhsi 470uf 16v - Bushytails comments

Originally Posted by bushytails

The 470uf 16v caps were Chhsi ... A quick DMM check showed they all had 5v on them (and from the board they looked like bypass caps), so just replaced them with the same 1200 @ 6.3 caps I used elsewhere.

I'm guessing I can raid my stash of Rubycon 820µF 16v ZLJ's left over from the A8N recap?

What I'm yet to sort are the last 5 - bushytails commented

Originally Posted by bushytails

1 Tayeh 1500uf 6.3v.
4 Teapo 1500uf 10v.

Whereas I simply have 5x Hitano 1500µF 10v (would have been the closest match from the local hobby shop in 2009).
Does anyone have notes on the layout of the board and what caps went where?
I'm kind of presuming the 4x 10v caps were the two pairs (EC1-2 & EC13-14) and the 6.3v one is the solo tucked in by the 6.3v VRM caps (labelled EC6). Does it particularly matter? Solder a couple of fly leads to the pins and run the multimeter across them to confirm?

Oh, and on my board, c390 was installed as per the silk-screen and not vented. Interesting comparison to Topcat's experience here Design Flaw on MSI 694D Pro ver:1.0
Assuming I get this puppy running again, I've got a couple of PCI serial cards spare that I might chuck in slots 2 & 3 to see what happens... or just attach a couple of fly leads and see whats what with the DMM?

Thanks all
Aeth..

[Aetherone]

Originally Posted by Aetherone

Oh, and on my board, c390 was installed as per the silk-screen and not vented. Interesting comparison to Topcat's experience here Design Flaw on MSI 694D Pro ver:1.0

Fly leads attached, -3.28v on that one. Topcat was right, no idea why mine had not vented with years of accumulated runtime. No adjacent PCI cards perhaps?
That one will get flipped around in the morning. I presume a few minutes of reverse polarity runtime will not have significantly affected the cap?

I also tacked fly leads to EC6 pins which gave 200 millivolts???
I was expecting a bit more or am I barking up the wrong tree without an oscilloscope?

[Topcat]

Originally Posted by Aetherone

Fly leads attached, -3.28v on that one. Topcat was right, no idea why mine had not vented with years of accumulated runtime. No adjacent PCI cards perhaps?
That one will get flipped around in the morning. I presume a few minutes of reverse polarity runtime will not have significantly affected the cap?

I also tacked fly leads to EC6 pins which gave 200 millivolts???
I was expecting a bit more or am I barking up the wrong tree without an oscilloscope?

Always love seeing classics revived!

[momaka]

Originally Posted by Aetherone

Does anyone have notes on the layout of the board and what caps went where?

Post a top-down picture of the board in a decent resolution, and we may be able to guess/check what's what.

I usually make my own "cap maps" with nothing more than a multimeter testing in resistance mode. Like this:
https://www.badcaps.net/forum/attach...1&d=1498712430
(Though now I make them on the computer, like this.)

First, I make note of which caps' positive leads are attached to each other, then map out the caps directly connected to the PSU's 3.3V, 5V, and 12V rails. After that, it's figuring out which one is for the CPU V_core, CPU Vtt (socket 754 and 775 or newer boards), RAM Vdd, RAM Vtt (for DDR 1/2/3/4), chipset/NB, SB, and etc. Then finally marking what's what on my pictures or hand-drawn diagrams.

Originally Posted by Aetherone

Fly leads attached, -3.28v on that one. Topcat was right, no idea why mine had not vented with years of accumulated runtime.

Wow, a cap in backwards from the factory by design? That's quite a goof there on MSI's part.

I imagine it didn't bulge because it was a 16V cap? And the voltage was not too high (only 3.3V).

Most electrolytic caps can handle a very small amount of reverse voltage. The lower the reverse voltage relative to the cap's rating, the less damage that will occur to the cap. But damage *will* occur. It's just that with a very small reverse voltage, the damage may be fairly slow, allowing the cap to vent its gases slowly through the bung over time... and thus never bulge. Eventually, it may become high ESR / open-circuit and thus not care at all about the reverse voltage.

Originally Posted by Aetherone

I also tacked fly leads to EC6 pins which gave 200 millivolts???
I was expecting a bit more or am I barking up the wrong tree without an oscilloscope?

Just use your multimeter on resistance mode and see what rail that cap goes to. It may be on the PSU or it may be for something else.

HINT: remove CPU, RAM, add-in cards, CMOS battery, and disconnect PSU when doing these tests. Otherwise, all of these may give you a false reading.

HINT 2: large caps and groups of parallel caps may take a while to charge up when resistance is being tested with a multimeter. This can make you conclude that two or more rails are connected to each other when they aren't. To prevent this false reading, allow 3-4 seconds for the multimeter to charge any caps before accepting the resistance reading.

HINT 3: use lowest resistance setting if using a manual multimeter. Readings below 3 Ohms typically indicate two test spots are connected to each other. And readings between about 5 Ohms and 50 Ohms are typically chipset/NB/SB static off resistance. (So really, anything over 5 Ohms, we don't care about.)

[Aetherone]

IT LIVES!

Well, what a change a new set of caps makes. She seems to be ripping (lol) along with no issues so far. Even the front USB header is working again.

All I need now is to find a suitable AGP card and its complete!

I've love to find a dual-DVI quadro card (workstation GPU for a workstation board right?!), preferably one that works better than the Quadro4 900XGL now sitting in my e-waste recycling box.

Also love to find a PCI SATA/SAS raid card to drive an old SSD. Not exactly period correct, but spinning rust is just horrifically slow at any RPM.

[momaka]

Good to hear you got it running!

Originally Posted by Aetherone

All I need now is to find a suitable AGP card and its complete!

Well, if you're deadset on having an nVidia workstation GPU again, look up Quadro FX 1000 (or alternatively "180-10192-0000-A01") or Quadro FX 2000. Both of these are based on the NV30 - i.e. same chip as the GeForce FX 5800. Sure, the GeForce FX series weren't nVidia's best in terms of performance (the high-end FX cards did only slightly better than the Ti 4400 and Ti 4600). But they were a bit more reliable and said cards above can typically be found for around $25 shipped (at least here in the USA on eBay), unlike the Quadro4 / GeForce 4 Ti cards, which sometimes go for outrageous sums right now. Just keep in mind that these Quadro FX 1000/2000 cards have a pretty loud fan.

Otherwise, if you want decent performance at lower price point, Radeon 9600 cards can often be found fairly cheap. Though most of those only come with one DVI port. So probably wiser to look for a newer ATI card. If super-high performance isn't of importance, something like the VisionTek X1300 should be pretty solid. Those use little power and run pretty cool. Same with the Radeon HD2400 AGP - but that's a 64-bit memory bus video card, and it will have terrible performance at memory-intensive loads.

The GeForce 6600 GT is another very very decent performing AGP video card... but its price can be a bit high now due to demand. An alternative, the Radeon x1650 AGP, should perform fairly similar.

Originally Posted by Aetherone

... preferably one that works better than the Quadro4 900XGL now sitting in my e-waste recycling box.

I assume that's due to artifacts, isn't it?
I have a bad TI 4600 like that. nVidia put too small of a cooler on the GPU chip, allowing it to run too hot and fail over time. So for anyone who owns one of these cards and they still work - swap that tiny reference nVidia cooler for something capable of handling 30 Watts TDP properly.

[Aetherone]

Originally Posted by momaka

Well, if you're deadset on having an nVidia workstation GPU again

Well it's mostly so it matches the dual-DVI Quadro NVS280 PCI that's currently driving its video. Quad head FTW!

Shipping to Australia is proving a bit of a sore point though. Since this project has been 18 years in the making, I might just put it on the backburner and keep an eye on ebay for a while.

Originally Posted by momaka

I assume that's due to artifacts, isn't it?

Yup, cooked to death. The teeny HSF was entirely lined with felt and the thermal paste had a ring of badly cooked & oxidised carrier wax baked to it.

[Sparkey55]

Originally Posted by momaka

Good to hear you got it running!


Well, if you're deadset on having an nVidia workstation GPU again, look up Quadro FX 1000 (or alternatively "180-10192-0000-A01") or Quadro FX 2000. Both of these are based on the NV30 - i.e. same chip as the GeForce FX 5800. Sure, the GeForce FX series weren't nVidia's best in terms of performance (the high-end FX cards did only slightly better than the Ti 4400 and Ti 4600). But they were a bit more reliable and said cards above can typically be found for around $25 shipped (at least here in the USA on eBay), unlike the Quadro4 / GeForce 4 Ti cards, which sometimes go for outrageous sums right now. Just keep in mind that these Quadro FX 1000/2000 cards have a pretty loud fan.

Otherwise, if you want decent performance at lower price point, Radeon 9600 cards can often be found fairly cheap. Though most of those only come with one DVI port. So probably wiser to look for a newer ATI card. If super-high performance isn't of importance, something like the VisionTek X1300 should be pretty solid. Those use little power and run pretty cool. Same with the Radeon HD2400 AGP - but that's a 64-bit memory bus video card, and it will have terrible performance at memory-intensive loads.

The GeForce 6600 GT is another very very decent performing AGP video card... but its price can be a bit high now due to demand. An alternative, the Radeon x1650 AGP, should perform fairly similar.


I assume that's due to artifacts, isn't it?
I have a bad TI 4600 like that. nVidia put too small of a cooler on the GPU chip, allowing it to run too hot and fail over time. So for anyone who owns one of these cards and they still work - swap that tiny reference nVidia cooler for something capable of handling 30 Watts TDP properly.

Thermaltake made an after market solid copper heatsink/fan that fit those cards very well. In reference to the Ti.

[Aetherone]

Originally Posted by Sparkey55

Thermaltake made an after market solid copper heatsink/fan that fit those cards very well. In reference to the Ti.

I've got a beautiful all-copper Zalman heatpipe flower tucked away but first I need a card that's not already cooked to death

[momaka]

Originally Posted by Aetherone

... and the thermal paste had a ring of badly cooked & oxidised carrier wax baked to it.

Was it yellow in color? If so, that's actually normal. During that era, both ATI and nVidia used this yellow thermal "gunk" / glue that hardened over time.

I usually scrape it off with a razor blade as much as I can. Then after that scrub with acetone-soaked pad. It's the only thing that will soften it and remove it. IPA doesn't seem to dissolve it at all.

Originally Posted by Aetherone

... but first I need a card that's not already cooked to death

Well, if it's cooked to death... you can always cook it to death again. By that, I mean reflow it. I'm not sure if that will work or not, though, as this is not a flip-chip GPU. But I suppose it's worth a shot anyways.

Because these GF4 TI cards have electrolytic caps, you CANNOT oven-reflow them (otherwise the caps will pop). Has to be with a heat gun / hot air station.

Of course, all of that is if you even want to mess with the card.

Originally Posted by Sparkey55

Thermaltake made an after market solid copper heatsink/fan that fit those cards very well. In reference to the Ti.

Yup.
Or just about any socket A or Pentium 4 heatsink will do fine too (maybe some Pentium 3 ones, as well, if they are bigger.) The GF4 Ti cards have about 30-35W TDP, which isn't that hard to keep cool... though certainly not possible with the tiny stock coolers, which are good for 15W TDP at most. Maybe 20W with cooler ambient temps. socket A and Pentium 4 coolers will handle the heat just fine. Only hard part is figuring a way to mount these coolers on the video card.

[Sparkey55]

Originally Posted by momaka

Was it yellow in color? If so, that's actually normal. During that era, both ATI and nVidia used this yellow thermal "gunk" / glue that hardened over time.

I usually scrape it off with a razor blade as much as I can. Then after that scrub with acetone-soaked pad. It's the only thing that will soften it and remove it. IPA doesn't seem to dissolve it at all.


Well, if it's cooked to death... you can always cook it to death again. By that, I mean reflow it. I'm not sure if that will work or not, though, as this is not a flip-chip GPU. But I suppose it's worth a shot anyways.

Because these GF4 TI cards have electrolytic caps, you CANNOT oven-reflow them (otherwise the caps will pop). Has to be with a heat gun / hot air station.

Of course, all of that is if you even want to mess with the card.


Yup.
Or just about any socket A or Pentium 4 heatsink will do fine too (maybe some Pentium 3 ones, as well, if they are bigger.) The GF4 Ti cards have about 30-35W TDP, which isn't that hard to keep cool... though certainly not possible with the tiny stock coolers, which are good for 15W TDP at most. Maybe 20W with cooler ambient temps. socket A and Pentium 4 coolers will handle the heat just fine. Only hard part is figuring a way to mount these coolers on the video card.

For any TIM that alcohol will not dissolve I use GooGone then clean up with denatured alcohol.

[Aetherone]

Originally Posted by momaka

Was it yellow in color?

Yep. Although I've never seen one quite so badly cooked before. The ring around the edge exposed to air was burnt brown!

Originally Posted by momaka

Well, if it's cooked to death... you can always cook it to death again.

Decapped, fluxed and cooked. No real change to the failure
Not a FCPGA but still has plenty of BGA solder balls to go wrong.

Originally Posted by momaka

Or just about any socket A or Pentium 4 heatsink will do fine too

Wayyyyy back in the day when GF4 were new and hot my shiny new Asus TI4600 lasted about three weeks with its stock sink before I customized a P3-1Ghz sink to go on it.