douglar wrote on 2021-05-13, 18:34:

Is the "9116" chip the ramdac?

no 9116 is the datecode, chip is JT82C176-66

and looking at the cirrus logic board. CL-GD5906 does look to be the rom chip, judging by the traces on the board.

weedeewee wrote on 2021-05-13, 19:01:

douglar wrote on 2021-05-13, 18:34:

Is the "9116" chip the ramdac?

no 9116 is the datecode, chip is JT82C176-66
and looking at the cirrus logic board. CL-GD5906 does look to be the rom chip, judging by the traces on the board.

I was hoping there was a chance that it was a ramdac for a second because it is close to the VGA out, but I see what you mean. The rom traces go to the 8bit portion of the ISA slot, while the VGA traces go through the ferrites and towards the back of the card. Looking at wikipedia, it appears that Cirrus integrated the ramdac into the main chip with the CL-GD5420, so there's no replacing it.

OK, thanks for the info.

@douglar
I think your cirrus logic has the ramdac integrated to the cirrus logic chip so you're out of luck with that one (maybe a recap would help ?)

@maxtherabbit
For the trident I think it's the same card with a different BIOS. Your bios is separated between two ROMs, a "LOW" and a "HIGH" one (notice the "L" and "H" on each of them respectively). On mine it's one, larger ROM (notice the "LH" printed on the chip meaning it combines low and high bytes).

Can you please test this card with pinball fantasies ? the ramdac is similar to the original one on the first trident I improved (it was DT branded). I want to see if it shows the same palette problems. (FYI, the game was configured in high res mode).

Deksor wrote on 2021-05-13, 21:35:

@douglar
I think your cirrus logic has the ramdac integrated to the cirrus logic chip so you're out of luck with that one (maybe a recap would help ?)

correct, the cirrus branded DIP package is the ROM

Deksor wrote on 2021-05-13, 21:35:

@maxtherabbit
For the trident I think it's the same card with a different BIOS. Your bios is separated between two ROMs, a "LOW" and a "HIGH" one (notice the "L" and "H" on each of them respectively). On mine it's one, larger ROM (notice the "LH" printed on the chip meaning it combines low and high bytes).

I know 🤣 that's why I specifically mentioned it had 16-bit ROM in my first post

Deksor wrote on 2021-05-13, 21:35:

Can you please test this card with pinball fantasies ? the ramdac is similar to the original one on the first trident I improved (it was DT branded). I want to see if it shows the same palette problems. (FYI, the game was configured in high res mode).

I'd be happy to, but I don't have that game - is it a shareware or what?

weedeewee wrote on 2021-05-13, 19:01:

douglar wrote on 2021-05-13, 18:34:

Is the "9116" chip the ramdac?

no 9116 is the datecode, chip is JT82C176-66

and looking at the cirrus logic board. CL-GD5906 does look to be the rom chip, judging by the traces on the board.

yes

douglar wrote on 2021-05-13, 18:34:

Are those ferrites on the card near the VGA out ? Do they contribute to the soft picture?

they are and they are an element of an LC filter on the output

I can try removing the filter and see if that sharpens it up. I'm more concerned by the fact that this, as well as a couple of other cards I have (another Trident 8900 and a WC90C31LR) have 150R pulldowns on the RAMDAC outputs instead of 75R. The datasheets I've been able to find for these DIP type RAMDACs all specify 75R pulldowns (which makes sense since VGA Z0 is 75)

@maxtherabbit Looks like a free trial of pinball fantasies is available: https://www.bestoldgames.net/pinball-fantasies

@Deksor When I had 2 ramdacs ( ATI68830 and TLC34076 ) installed on a mach 32 card, I would get the random pink colors showing up in the VGA palette (among other more pressing problems) Seems really possible that the pink colors you are seeing are a ramdac malfunction of some sort.

My trident card only has one Bios chip and a samsung ramdac.

It also has vertical banding that looks like yours.

Getting back to the mach32 card, I see this picture: http://www.vgamuseum.info/images/palcal/ati/mach32pcib.jpg

Makes me think that a BT481AKPJ110 PLCC-44 might work on that board. I'll see if I can do the swap.

maxtherabbit wrote on 2021-05-14, 01:00:

I'd be happy to, but I don't have that game - is it a shareware or what?

Well no it's a complete version I have (which is hacked). Get it the way you prefer. The great things about this game for testing VGA is that it works even on 16 bit CPUs and it's really fast. Also the weird high res mode it has seems to trigger some artifacts more than 320*200 mode.

@douglar check the datasheet of the chips. You'll see that they're pin compatible 😀

Deksor wrote on 2021-05-14, 07:57:

Well no it's a complete version I have (which is hacked). Get it the way you prefer. The great things about this game for testing VGA is that it works even on 16 bit CPUs and it's really fast. Also the weird high res mode it has seems to trigger some artifacts more than 320*200 mode.

Can your monitor tell you what signal it is getting the high res mode? Sometimes if you hunt through the on screen controls, you can find some info about the incoming signal.

Looks like your trident was an 8800B, which is an older version. I like how it has all the different oscillators on it instead of a single crystal & Trident tck9002np PLL chip .

How important is the PLL quality for output?

I don't know. What's a PLL actually ? I'm not really knowledgeable in electronics (this is one of the reason why I made this thread, to share my experience and hopefully have people more knowledgeable in electronics' opinion ^^)

Deksor wrote on 2021-05-14, 17:15:

I don't know. What's a PLL actually ? I'm not really knowledgeable in electronics (this is one of the reason why I made this thread, to share my experience and hopefully have people more knowledgeable in electronics' opinion ^^)

I remember PLL ( https://en.wikipedia.org/wiki/Phase-locked_loop ) from the days when you had to change a crystal to change your motherboard speed. A PLL takes in one clock speed from a crystal and can spit out several anothers, so that you can easily change the mobo frequency with jumpers or later in the bios. It was definitely a feature you wanted on a hobbyist motherboard, so I learned to look for those back in 1994.

Looks like your Trident VGA card is older and it has different crystals for each frequency, while the newer tridents have a single crystal and a PLL to generate different refresh rates.

douglar wrote on 2021-05-14, 17:29:

Looks like your Trident VGA card is older and it has different crystals for each frequency, while the newer tridents have a single crystal and a PLL to generate different refresh rates.

Cirrus GD-542x series also uses a PLL with a 14.318Mhz input to generate its pixel and memory clocks. (This was chosen because a 14.318Mhz clock is available "for free" on the ISA bus.)

It's a "fractional N" PLL, so if I understand correctly it first multiplies the input clock by a configurable amount M and then divides it by a configurable amount N to produce the final frequencies. As a result the card is never going to hit the exact VGA/SVGA clocks, but the result will be within 1-2% of the correct frequency.

At one point I was reading through this card's driver code from XFree86 and there were comments indicating that certain M/N combinations weren't stable or could cause interference. (Especially if the pixel and memory clocks were close to each other.)

I think SVGA cards that use a PLL to generate the clocks will look bad on a modern digital display. My theory: the sampling clock in your display is *more accurate* than the card's PLL. So the digital display is expecting VGA pixels to arrive at exactly 25.175Mhz but the card's PLL is producing pixels at say 25.181Mhz. This small difference obviously doesn't matter one bit to a CRT whereas a digital display might end up trying to sample a 'pixel' at the same moment that the card's RAMDAC is switching between two pixel values.

douglar wrote on 2021-05-14, 17:29:

I think SVGA cards that use a PLL to generate the clocks will look bad on a modern digital display. My theory: the sampling clock in your display is *more accurate* than the card's PLL. So the digital display is expecting VGA pixels to arrive at exactly 25.175Mhz but the card's PLL is producing pixels at say 25.181Mhz. This small difference obviously doesn't matter one bit to a CRT whereas a digital display might end up trying to sample a 'pixel' at the same moment that the card's RAMDAC is switching between two pixel values.

Actually, I've found 542x cards quite compatible with my Samsung 19 inch LCD. Multiple cards, no problems that I can recall.

It really is a card to card thing ... but Tridents seem very commonly bad!

So in that case does that mean that adding a proper 25.175MHz crystal to the circuit and bypassing the PLL could improve the quality ?

Guys - the sampling problems on modern displays are NOT due to tolerance problems in old hardware. They are by design

maxtherabbit wrote on 2021-05-03, 15:59:

Correct, but the problem is not necessarily that the cards are outputting an off-spec signal. The pixel clocks between 720x400 (VGA Text mode) and 320x200 (double-scanned low res graphics mode) are different despite the line counts being identical. Nearly all LCD monitors sample at 900px/line to get sharp text in text mode, but you then end up with jailbars on DOS games which use the low res graphics mode - since it should be sampled at 800px/line and the monitors have no way to distinguish which mode is being used due to identical line counts.

I wonder what results one could get by running the output of an old card with crappy output quality on an LCD through an OSSC or Extron DVI 300 . Both of these bypass the VGA text/graphics mode sampling issue, but beyond that, I am curious as to how good a result one could get without having to modify the old cards themselves (especially with an OSSC) and possibly some additional hardware. Obviously, some things, like bandwidth constraints on a cards analogue output stage cannot be fixed externally and impedance mismatches would likely require a tunable impedance matching circuit of some kind, but being able to manually adjust timings precisely to match whatever an old card is spewing might be enough to improve results significantly .

maxtherabbit wrote on 2021-05-15, 14:22:

Guys - the sampling problems on modern displays are NOT due to tolerance problems in old hardware. They are by design

maxtherabbit wrote on 2021-05-03, 15:59:

Correct, but the problem is not necessarily that the cards are outputting an off-spec signal. The pixel clocks between 720x400 (VGA Text mode) and 320x200 (double-scanned low res graphics mode) are different despite the line counts being identical. Nearly all LCD monitors sample at 900px/line to get sharp text in text mode, but you then end up with jailbars on DOS games which use the low res graphics mode - since it should be sampled at 800px/line and the monitors have no way to distinguish which mode is being used due to identical line counts.

I have some cards that produce jail bars at 320x200 and others that don't on the same monitor, cable, input port.

Are you just saying that the ones that make jail bars are a result of the board design or the monitor design?

douglar wrote on 2021-05-17, 02:50:

I have some cards that produce jail bars at 320x200 and others that don't on the same monitor, cable, input port.

Are you just saying that the ones that make jail bars are a result of the board design or the monitor design?

I can't explain that one. Maybe the cards that don't show them are just so blurry that it masks the problem? Or maybe we have two different types of jailbars in mind

maxtherabbit wrote on 2021-05-17, 13:50:

I can't explain that one. Maybe the cards that don't show them are just so blurry that it masks the problem? Or maybe we have two different types of jailbars in mind

Here are two examples-- Same exact everything except for the video card.

A Quantum Designs CL-GD5429 VLB

Has clear, distinct pinstripes

While the STB CL-GD5426 VLB

Has a perfect picture:

Results are identical on all 3 VLB motherboards that I have:
* LION NiCE Green VL
* Shuttle HOT 419 v1.0
* FIC 486 VIP

Edit -- I uploaded the correct "pinstripe" picture

Are these photos of an LCD or are you using some kind of capture?