I'd poke around to find out if the other chips BA0/1 signals are connected together (not sure if these are bank select or what), though it might be some chips and not all. If the signals are shared, and you can figure out with which chips, it might just work with some extra wires. Keep in mind though at the speeds these signals switch the wires might just not cut it, in which case if you can't somehow extend the existing traces to the missing pads the card is toast.

Deunan wrote on 2024-07-04, 10:29:

I'd poke around to find out if the other chips BA0/1 signals are connected together (not sure if these are bank select or what), though it might be some chips and not all. If the signals are shared, and you can figure out with which chips, it might just work with some extra wires. Keep in mind though at the speeds these signals switch the wires might just not cut it, in which case if you can't somehow extend the existing traces to the missing pads the card is toast.

The chips on the other side seem like they have linked BA0/1, so I guess on the GPU die side will be linked too. There are some traces left from BA0/1, but I don't have enough skill to solder on replacement solder pads (they are going under the chip itself). If I where to connect the two pins with wires, would I get artifacts? I think these pads are bank select, at least from what I can understand.

Try running wires, as short as you can make them but also try to keep them both the same length, from a nearby chip that you suspect shares the same signals with the one that has broken pads. Will it work, well I don't know. For slower signals it should but the remaining, now unconnected traces will cause some signal reflections. You simply won't know unless you try, and these wires can always be removed if it just doesn't cut it.

Broken pads happen, especially on older PCBs that got pretty hot and/or went through many heat cycles. Wort case scenario you'll get some soldering experience out of it.

Managed to solder the wires to the correct pins, somehow. Can't say if the artifacts in DOS mode got worse or better since my TV has support problems for VGA, but in windows I was greeted with even worse artifacts (picture in attachments). Now it won't change the resolution and i see some glitching when moving the mouse cursor. I think its gonna be RAM connection issues or damaged GPU core. Any ideas?

kotel wrote on 2024-07-05, 13:16:

I think its gonna be RAM connection issues or damaged GPU core. Any ideas?

It could still be the RAMs, this is not a simple VGA where all the logic is on the ASIC and the RAM is only used as pixel buffer. With drivers loaded there might be parts of card RAM assigned to some back buffers, etc.
First I would try disconnecting the wires, just one at a time, to see what changes (if the system still boots up in that state). In case you picked a wrong chip to provide the signals. Obviously double-check the soldering for any shorts.
Second you said there is a 3rd pad but seems unused. Did you check if that pad does seem unused also on other chips? I mean the trace might be hidden so you can't be sure, but if you see the pin connected to something on other chips then perhaps this missing one also was connected somewhere.

In general vertical lines like this would suggest addressing issue, or one particular chip of many being dead or having faulty output. You can't change the resolution, how about color depth? Did you try switching between 256, 32k/64k and 16M color modes? Since the RAM chips will carry different pixel data the vertical line patterns should also change a bit. 256-color mode, with one byte per pixel, might be most interesting to see.

Deunan wrote on 2024-07-05, 13:31:

It could still be the RAMs, this is not a simple VGA where all the logic is on the ASIC and the RAM is only used as pixel buffer […]
Show full quote

kotel wrote on 2024-07-05, 13:16:

I think its gonna be RAM connection issues or damaged GPU core. Any ideas?

It could still be the RAMs, this is not a simple VGA where all the logic is on the ASIC and the RAM is only used as pixel buffer. With drivers loaded there might be parts of card RAM assigned to some back buffers, etc.
First I would try disconnecting the wires, just one at a time, to see what changes (if the system still boots up in that state). In case you picked a wrong chip to provide the signals. Obviously double-check the soldering for any shorts.
Second you said there is a 3rd pad but seems unused. Did you check if that pad does seem unused also on other chips? I mean the trace might be hidden so you can't be sure, but if you see the pin connected to something on other chips then perhaps this missing one also was connected somewhere.

In general vertical lines like this would suggest addressing issue, or one particular chip of many being dead or having faulty output. You can't change the resolution, how about color depth? Did you try switching between 256, 32k/64k and 16M color modes? Since the RAM chips will carry different pixel data the vertical line patterns should also change a bit. 256-color mode, with one byte per pixel, might be most interesting to see.

I don't think the system wouldn't boot up without those wires since the card was working without the pads, although it had artifacts too, but not the same type. I don't like to desolder those wires since it was too hard to even get them to stick on. I checked for short and found none. Also checked if the NC pad on other chips is used. Not to my surprise it wasn't. Mind telling me how to change the color depth on XP? Couldn't find it anywhere. In the meantime I'll try to find an spare VGA monitor so I could run MONTEST.

kotel wrote on 2024-07-06, 14:56:

Mind telling me how to change the color depth on XP? Couldn't find it anywhere.

Right-click on the desktp, select Properties, then the (last) tab Settings. It's right next to resolution setting.
That being said, now that I look at your photo again I think it's just 16 colors? Stuck in VGA mode in other words, perhaps the driver did not load due to damaged RAM (this could prevent size detection for example).

Yes, I realize that soldering those wires can be a major PITA and you'd rather not touch it again. But if nothing else works, what other options are there...

Deunan wrote on 2024-07-06, 15:38:

Right-click on the desktp, select Properties, then the (last) tab Settings. It's right next to resolution setting. That being sa […]
Show full quote

kotel wrote on 2024-07-06, 14:56:

Mind telling me how to change the color depth on XP? Couldn't find it anywhere.

Right-click on the desktp, select Properties, then the (last) tab Settings. It's right next to resolution setting.
That being said, now that I look at your photo again I think it's just 16 colors? Stuck in VGA mode in other words, perhaps the driver did not load due to damaged RAM (this could prevent size detection for example).

Yes, I realize that soldering those wires can be a major PITA and you'd rather not touch it again. But if nothing else works, what other options are there...

Rebooting the system finally fixed that I couldn't change resolution and color modes. Got into DOS on an monitor with broken screen. Pictures of color tests from MONTEST in attachments (the black spots on the screen are monitors fault). Also did memory tests. Tests that passed are CGA and VGA. Tests that failed are text mode, EGA and planeswitch, VESA and bankswitch. Also attaching pictures of color output tests with picture of windows in 1400x1050 32 colors without removing the cables.

I tried some pixel counting and the 320x200 screenshot seems to have vertical lines 56 pixels tall. Which is a bit odd number if we assume exactly one RAM chip is bad. And every second column as well. But as I've said these chips don't really map VRAM to screen in as simple way as old VGA would.

There is a pattern there, 8 color bars, which repeats. So I'm not convinced this is actually coming from RAM, perhaps it's more of a floating bus that retains part of the address. Are these lines stable? Or do they change quickly?
Look into datasheet for these chips, I would assume that bank lines are tied together for a group of chips then the address and other controls lines would also be like that. Except data lines, those should have all independent connections to the video chip (except maybe can be shared between the banks, so probably between the chips on the opposite sides of the PCB). Check if all the other lines are also connected to other chips, perhaps the one with now damaged pads actually has more pins needing a wire then you thought.

Deunan wrote on 2024-07-08, 19:24:

I tried some pixel counting and the 320x200 screenshot seems to have vertical lines 56 pixels tall. Which is a bit odd number if we assume exactly one RAM chip is bad. And every second column as well. But as I've said these chips don't really map VRAM to screen in as simple way as old VGA would.

There is a pattern there, 8 color bars, which repeats. So I'm not convinced this is actually coming from RAM, perhaps it's more of a floating bus that retains part of the address. Are these lines stable? Or do they change quickly?
Look into datasheet for these chips, I would assume that bank lines are tied together for a group of chips then the address and other controls lines would also be like that. Except data lines, those should have all independent connections to the video chip (except maybe can be shared between the banks, so probably between the chips on the opposite sides of the PCB). Check if all the other lines are also connected to other chips, perhaps the one with now damaged pads actually has more pins needing a wire then you thought.

On all of the color testing pictures they are stable. In BIOS, they are flashing and there are also color stripes like in the pic of windows in 16 color mode. Checked to see if the other BA1/0 are connected to all of the chips, and they are. Rechecked my connections around 10 times to see if I connected them right (didn't want to kill another GPU and damage another mainboard). Address input pins seem to be connected only with chips on the same side. DQ (data I/O) isn't connected to other chips. Any ideas?

After some thinking, the every 2nd column being bad is perhaps due to the card alternating between the front side and back side chips. That doesn't explain the odd pixel count but I'm pretty sure I won't be able to guess this unless I know more about how this chip operates low-level, and I don't have the time to study this even if I could find some nice docs (going through Linux sources of this complexity is not a pleasant experience).

So if it's stable and not randomly blinking, and also the lines seem to have uniform color, I would say the bus is floating during the read. On these RAMs address and data are not shared but perhaps on the GPU side there is some cross-talk. If the chip was working but just the wrong bank was selected I would expect to see either random but static data (uninitialized part of RAM) or wrong data being shown. Not semi-random colored lines. Although on some of the screenshots it kinda looks like parts of the screen are just in wrong places, but there are still weird vertical lines.

Next to BA0 and BA1 lines there are also control signals, like /CS for example. Did you check if those are also shared across the RAM chips on each side of the PCB? I would assume they are. Make sure you are not missing any of those on the suspect chip. Your options are the chip is not really responding to reads at all, or the bank select is badly wrong and the colored lines are in fact what's in the chip (perhaps a left-over from some initialization / memory count procedure). But it's not possible to tell without a scope or more experiments with the BA lines unconnected.

If you don't feel confident resoldering the wires again then just stash the card somewhere, return to it after a few years when you get more soldering experience. There is no rush here I think? Because the other choice is to assume the card is beyond repair now and only good for parts or a trash bin.

Deunan wrote on 2024-07-09, 11:14:

Next to BA0 and BA1 lines there are also control signals, like /CS for example. Did you check if those are also shared across the RAM chips on each side of the PCB? I would assume they are. Make sure you are not missing any of those on the suspect chip. Your options are the chip is not really responding to reads at all, or the bank select is badly wrong and the colored lines are in fact what's in the chip (perhaps a left-over from some initialization / memory count procedure). But it's not possible to tell without a scope or more experiments with the BA lines unconnected.

All /CS lines are connected together. I suspect that, when I desolder those BA cables, I should see in DOS screens white lines and flashing cyan/blue lines. Will try to desolder those cables after I find some time.

If you do desolder the wires, try to desolder them at the good chip side. With the wires still attached to the pins with missing pads. This way you can connect them both to some nearby GND point (add extra wire if need be, this is not so critical if the signal is stable and never changes). This will force the suspect chip to always use bank 0, leaving the rest working as-is. Some chances are DOS mode for the card will not use banks higher than zero so this would at least work until drivers are loaded. If it does work that way then you'd at least know the chip is OKish, just the bank switching is not working.

BTW did you take a good look at the missing pads? Is there a chance the signals were routed through the pads and going to next chip? Becuase if these are daisy-chained and the connection is now missing you might need more than these 2 wires to fix it.

Deunan wrote on 2024-07-09, 15:16:

If you do desolder the wires, try to desolder them at the good chip side. With the wires still attached to the pins with missing pads. This way you can connect them both to some nearby GND point (add extra wire if need be, this is not so critical if the signal is stable and never changes). This will force the suspect chip to always use bank 0, leaving the rest working as-is. Some chances are DOS mode for the card will not use banks higher than zero so this would at least work until drivers are loaded. If it does work that way then you'd at least know the chip is OKish, just the bank switching is not working.

BTW did you take a good look at the missing pads? Is there a chance the signals were routed through the pads and going to next chip? Becuase if these are daisy-chained and the connection is now missing you might need more than these 2 wires to fix it.

Thanks for the additional info. All I can remember is that, both BA pads were going to a VIA. There is a slim chance I could try to solder on some replacement pads since there was a fair bit left of the trace, but I can't check if the ones I found are correct. Probably I'm gonna need to desolder this chip again since I didn't clean off excess rosin flux (or whatever you call it) and now the chip isn't laying on the PCB correctly, but enough to make some contact.

Here is the link to the google drive folder with pictures of my testing with the card since the folders exceeded maximum file size. Let me know if it doesn't work. I'll try to solder those two cables to some GND pad I can find somewhere close. What I didn't expect is that, after desoldering those two cables, there wasn't a screen with white and cyan stripes across. So I think there's some non connected pads left.

It's odd that without BA0 or BA1 the lines on 256 color test are breaking up, but with both signals unconnected they are solid again. But it could be due to some stray signals affecting the unconnected inputs. There is a difference on the Windows screen, and both signals missing makes it worse but still the lines persist as they were. So I would concur that some other signals are not reaching the chip. If you do remove the chip again try making a good photo of the pads, might help you later on.

You might find this interesting, I've recently bought some very cheap (~3 EUR each) AGP cards to run some benchmarks and output quality for DOS games. One of those is Radeon 7000 which seems to work OK in DOS text mode but in graphics modes I see some of the patterns of your 9600:
- In 320x200 I have the 2 rows of vertical lines, but not as dense as yours (every 8 pixels or so) and they seem to be like a shift of the correct color rather than repeating color bars. For example black is unaffected.
- In VESA 640x480 I get vertical lines all the way down, but in games where the pages are flipped for animation there is some pixel dancing visible in parts of the lines. Top of the screen is more active and the bottom almost static. As if it depended on the address of the page in memory.
This is without drivers because I don't have a Windows testbed prepared for them yet. Pure DOS 6.22.

I suspect a bad chip connection, or a damaged RAM chip. I will investigate it further and let you know if I come to any conclusions.

EDIT: Actually, I found the problem already. This will be tricky to repair since I don't have spare DDR chips and one of leads looks to be completly separated. But I might be able to remove some of the plastic and solder a wire, if it's only the middle lead of the 3 that's completly cut. I think the plastic removal is going to be the most grueling work...

Looking through my donor card's box I found an Geforce FX5500 with a fried core which has the same memory as the 9600se, so I suppose I could try to swap the chips and see if there's any difference. I'll try to desolder the chip with missing pads and take a picture of pads and where they connect.

Well I got the chip connected, soldering the wire to the stub was actually pretty difficult. I didn't want to remove more plastic because that could expose nearby pins as well, and cause even more problems during soldering. I'm not too happy with this wire but it's good enough.

Turns out however this wasn't the only damage, one of the 10 ohm resistor packs is missing a quarter of it. Broken clean off, and thus another bit is unconnected. I think I can just put a 0402 resistor in there and not have to redo the whole pack (I hate manual soldering these things). But I don't have spares on hand, ordered some today. I will make one last update when I have that repaired as well - I remain optimistic though.