Speaking of elsa winner, and sorry for being offtopic but how does it use the eeprom of it? I mean, what bits and type of the eeprom?

Looks like video RAM bit errors. Sometimes caused by insufficient decoupling, sometimes by bad mobo or PSU, sometimes by some actual defects.

Edit: Try it in other computers. I have two otherwise very reliable Trio64v2 cards that show such errors in bad computers.

Thanks for the answer. As all other cards were happy on the same computer but this one I was pretty sure it was the card, but it's never a bad test so good call.

Here are the results.

1. on a Gigabyte 440BX mobo - funnily letters are fine there, but the BIOS display on boot is all corrupted (too quick to capture) and there is still corruption on the EPA logo, on the award logo top-left, and the whole left side.

2. on a DELL P4 mobo - looks very much like the same kind of corruptions as the Socket 7 mobo pictures in my first message.

The display on the Dell mobo - I love it, it looks so Welsh 😀

I'd really love to put the PSU(s) being used onto my test bench and take some measurements.
Many AT (and some old ATX) PSU's don't respect the +-5% tolerance for +5V, many drop to 4.5V under load, I even encountered some that drop down to 4.2V to prevent the +12V going too high.
In such cases there are always some devices that are more sensitive and show malfunctions.
To my feeling, the Trio64V reference design seems to be particularly susceptible.

My personal guess: The card is perfectly OK, and the problem is the PSU.
Do you have a multimeter?
You could stick its "needles" into the AT/ATX power connector while the computers run and find out the voltages.

retardware wrote on 2021-09-08, 13:48:

The display on the Dell mobo - I love it, it looks so Welsh 😀

can't unsee...definitely looks like DOS welsh edition 🤣

I'd really love to put the PSU(s) being used onto my test bench and take some measurements. Many AT (and some old ATX) PSU's don […]
Show full quote

I'd really love to put the PSU(s) being used onto my test bench and take some measurements.
Many AT (and some old ATX) PSU's don't respect the +-5% tolerance for +5V, many drop to 4.5V under load, I even encountered some that drop down to 4.2V to prevent the +12V going too high.
In such cases there are always some devices that are more sensitive and show malfunctions.
To my feeling, the Trio64V reference design seems to be particularly susceptible.

My personal guess: The card is perfectly OK, and the problem is the PSU.
Do you have a multimeter?
You could stick its "needles" into the AT/ATX power connector while the computers run and find out the voltages.

Mmh well as said I'm definitely no expert of that kind of stuff so I happily take your word, but the Socket7 from my initial post has an "original" Enlight ATX PSU from back in the days, the 440BX board a modern Corsair PSU, and the Dell a Dell PSU (it's an original Dell Dimension 5150 case). I find it weird to believe that all 3 PSU's wouldn't be good enough for the card when I didn't have any issue with anything else thrown at those machines, empiricism would lead me to believe it's the card not working on all machines being faulty?

That said, I'm happy to learn, I do have a multimeter, so let's get rolling!

I never did that before, should I measure voltage between any red +5V pin and any black ground pin?

Yes, measure black-red (+5) and black-yellow (+12).
It is always useful to check +12V too.
Because, if it is really high, then it indicates that the loading ratio is out of balance for the particular PSU (too much load on +5, too little on +12V).

Only did the Dell so far as it's the most accessible. Results:
- 5.07
- 11.85
Looking good?

Well, this only rules out undervoltage.
The worst PSU I ever encountered had very precise voltages like these on the multimeter, but on the test bench with the oscilloscope it revealed 2Vpp ripple (Leaky caps).
Judging from the introduction date, the Dell PSU well might have been manufactured in the Age of Bad caps (~2000-2005).
So, personally, I'd not yet be convinced that the Trio64 is actually so much over borderline that should be called "bad".

Enlight: 5.08/12.30
Corsair: an impressive 5.00/12.00

OK so let's assume the card is bad, because everything points to the card having a problem, is there anything one could do to try and save it? 😁

Maybe it might be worth a try: slapping on some extra decoupling caps.
This often helps stabilize borderline boards/cards when overclocking.
But, that effort for a card that can be bought for ~10 euros?

So, I'd consider another variant more upgrade-worthy, for example this:
this

Less chaotic, plenty of empty places where you can add caps if needed.

Heh thanks but no I'm not looking to buy something else..I was seeing it more as a real use case to learn something new in diagnostic and maybe some easy soldering if there was a point 😀 but I have way enough cards already 🤣

Pleas press on DRAM (two below EPROM) and power it -> did change it ?

If no -> next step: Press on QFP IC (S3 Chip). Did change it `?

It is 100% DRAM Problem , causes damage DRAM or bad solder joint or bent lead. (Your card has many scratch, i wouldn't wondered, if PCB trace is cutted by scratch. )

regards
matt

Good call! Indeed, pressing the DRAM reaaaaal hard solves the issue temporarily!

Should I try to heat the chip legs with the heatgun?

Bake it!

honestly i'd just love to do that and is maybe the sole motivation of this whole thread 😁

so I lost the cooking instructions... would 10 minutes at 200°C fan-assisted tender it just right?

No, don't bake it. It doesn't help.

re-soldering DRAM. You need flux and good solder iron with chisel tip.
Only soldering help it.

regards
matt

Definitely don't bake it. Last time I baked Gigabyte GeForce 6600 GT the heat virtually destroyed capacitors. They were ok before. After they were all either bulging or leaked out. Solder melting point wasn't reached, I have an oven with digital temperature setting and I believe I used 200'C setting. The card was gradually heated and the cooled down. I ended up keeping the heatsink and throwing the card into rubbish bin.

Boohyaka wrote on 2021-09-08, 10:48:

Hey folks, […]
Show full quote

Hey folks,

I have this Elsa Winner 1000 PCI card that produces these interesting text corruption and problems as soon as it starts. Computer doesn't complain, no beep at all. Computer is fine with other PCI gfx cards, so the problem is with this card.
I have dumped the BIOS and compared it a known good version found online - identical.

Elsa1.jpeg

What do you guys make of it? I am in no way a hardware expert, I've visually checked it carefully though and it seems in pretty good and clean condition and don't see anything obvious.
Is this a case where baking it at 200°C for 10 minutes as I've read in some places would be worth a try to "refresh" solder joints, even if they don't look bad to me?
I've read a heatgun would be preferred in some cases, I do have a basic heatgun and no idea how hot it gets, what would be the proper way for me to do that?

Obviously these cards are pretty available and inexpensive (still very good, highly compatible DOS cards!) so I could really just drop it in the bin. But if I can salvage a bit of retro history and learn something new along the way, I'm definitely game! And in that case no pressure of ruining something of high value 😁

I'm attaching a picture of the card here and can provide more/better pictures of the back or other areas on request, if that can help, cheers!

Elsa2.jpeg

The bios chip is corrupted. You Can replaced it? I Can programmer new bios.....

Yep, you need to reflow the solder joint of the memory module that you pressed which made the display ok.

Baking the card won't do that, it will just cook it (which is not what you want to do)

A good soldering iron, flux and maybe a touch of solder will fix that memory module right up in a fraction of the time you would spend you would spend baking it. And would actually fix it, rather than just cook it (seriously, it isn't dinner, it doesn't belong in an oven!)