So it looks like this card hangs the system during writes, or maybe I/O access, if DOS prompt seems to work fine.

See if you can find some programs, other than games, like demos or benchmarks, that do a lot of graphics card access. In other words the contents of the screen must keep changing or be refreshed often. Doesn't need to be in graphics mode, text mode is fine too. In fact a hang in text mode would indicate a problem with writing to VRAM rather than I/O, since there is usually little to none I/O done for text.

The problem, as I see it, is there aren't really all that many chips on this card that can be causing this. Other than the two ASICs there is just a handful of 74LS chips. But those do die in weir way, being bipolar based. With good soldering equipment and skills it might be worth socketing and replacing those. The '244 near monitor connector can stay, all it should do is drive the video signal out. The others are potentially suspect. I would leave the VRAMs be for now, yes one of these might be very slow during reads and hogging the bus but that is rare. And to test that all the chips need to be replaced with a known good set anyway because such dynamic issues are next to impossible to diagnose otherwise.

And obviously look for any traces that are scratched and could possibly be higher resistance or breaking completly due to heat and PCB expansion. Other than semi-broken vias these problems are also quite difficult to locate.

Deunan wrote on 2026-01-24, 09:55:

So it looks like this card hangs the system during writes, or maybe I/O access, if DOS prompt seems to work fine. […]
Show full quote

So it looks like this card hangs the system during writes, or maybe I/O access, if DOS prompt seems to work fine.

See if you can find some programs, other than games, like demos or benchmarks, that do a lot of graphics card access. In other words the contents of the screen must keep changing or be refreshed often. Doesn't need to be in graphics mode, text mode is fine too. In fact a hang in text mode would indicate a problem with writing to VRAM rather than I/O, since there is usually little to none I/O done for text.

The problem, as I see it, is there aren't really all that many chips on this card that can be causing this. Other than the two ASICs there is just a handful of 74LS chips. But those do die in weir way, being bipolar based. With good soldering equipment and skills it might be worth socketing and replacing those. The '244 near monitor connector can stay, all it should do is drive the video signal out. The others are potentially suspect. I would leave the VRAMs be for now, yes one of these might be very slow during reads and hogging the bus but that is rare. And to test that all the chips need to be replaced with a known good set anyway because such dynamic issues are next to impossible to diagnose otherwise.

And obviously look for any traces that are scratched and could possibly be higher resistance or breaking completly due to heat and PCB expansion. Other than semi-broken vias these problems are also quite difficult to locate.

This would be much easier than finding another chips 😀 , changing LS's i meant. Will try to do this, but first i need to download any benchmark tools....

butjer1010 wrote on 2026-01-23, 17:39:

This is the screen when it won't work 😀

Oh, this picture is actually very interesting. Yesterday I drafted an elaborate response how this might look like video memory address bits leaking into video data bits, but the specific pattern you are seeing does not match the EGA architecture at all, but I lost that draft response when inadvertantly closing the browser window, and I re-considered whether this information is relevant to the thread at all, and decided to not re-write it. Today, I read more details of the P82C435/P82A436 data sheet, and actually the pattern very much matches what would happen if the memory data bus A is not driven by the RAM chips during CRTC reads! The P82C435 multiplexes memory addresses and data over the same pins (which is not done at all in the original IBM EGA design), and the character and attribute codes seen in that screenshot exactly match the row addresses driven on the multiplexed pins before the read happens. On the other hand, font reads work perfectly (they happen on memory bus B).

In 80-character text mode, the EGA scan-out logic cards needs every opportunity to access video memory, and fetches a character code and a attribute byte (on bus A) for one character while fetching the pixel data for the previous character (on bus B) all at the same time. This is true for both the original IBM EGA card and the 82c435 integrated EGA controller. As all characters and attributes are wrong, while all font accesses are working, the issue can not be a mismatch between the "high/low CRT bandwidth setting" and the divide-by-2 setting, which could be a symptom if the mode is incompletely programmed into the EGA card. Furthermore, the signal to enable reading of character codes on memory bus A is the same signal as the one used to enable reading font data on memory bus B, it's /OE02 on pin 23 of the 82c435, and as font reads obviously work, the signal /OE02 obviously works, too. I don't see a reason why a card designer would want to add logic that prevents /OE02 to reach character memory (and /OE13 to reach attribute memory), but still allows /OE02 to reach font memory. So maybe the theory of missing /OExx on memory bus A is wrong? Except for /OE, there are two other signals that need to be active for attribute and character memory to respond, which are /RAS and /CAS. If any of /RAS, /CAS or /OE are missing on memory bus A, the screenshot you showed could be explained, but all of these signals are shared between memory busses A and B. So this seems like a dead end.

On the other hand, I have no idea what the empty positions U18..U25 are supposed to be for, as well as the logic chips U7 to U9. The datasheet of the EGA chipset claims that the chip count for an EGA card based on the 82C435/82A436 is just 13, which will be the two core ASICs, the BIOS chip, the monitor output driver chip U3 and the multiplexer chip U2, as well as 8 RAM chips. So the extra logic chips are something STB added to this design for some unknown purpose. Taking an educated guess, it looks like U18-U25 can take another 256KB of video memory for a total of 512KB. also, X1 and X3 could take extra clock generators for modes like 640x480 or 800x600. The EGA chipset used on that card is fast enough to handle those modes if given a sufficiently fast clock. It thus seems your PCB is designed for Super EGA operation, including the option of 512KB, which would provide two pages of video memory even at 800*600, yet only the chips required for standard EGA operation are provided. The three extra logic chips are used for video memory bank selection, and suddenly stuff starts to make some sense. What if memory bus A and memory bus B can be switched individually between the populated standard RAM or the "missing" extension RAM? If memory bus A would be switched to the extension RAM, the screen shot would be perfectly explainable! U8 is a dual flip-flop chip, which can be used to store two bank select bits. U9 is a quadruple OR chip, which can be used to gate four active-low signals (like /RAS or /CAS). This means U8/U9 are perfectly suited to implement bank switching. I can't easily guess what U7 (a 6-bit inverter) might be used for, though.

So, in essence, this means that the broken text mode picture you show would be explainable by a perfectly working EGA card with the bank selection not being properly initialized. It would be easy for the card designer to connect the CLEAR or PRESET pins of the 74LS74 to some RESET signal (e.g. the one on the ISA bus) to make sure the card starts with bus A and B switched to "standard memory", and I would expect this card to be implemented that way. So the "broken text" picture you show might point to a fault in the bank switch logic, and in that case, it is likely not related to Prince of Persia freezing. Well... Are you sure the computer freezes, or does only the image freeze, and you can continue playing blindly? If you can continue blindly, maybe the bank switching logic inadvertantly switched to a mode in which the computer starts updating the extension memory (which isn't there), while the card still displays the standard memory (which will no longer be updated). If that is the case, the "freezing" might also be due to a fault in the bank switching logic!

So indeed, U7 to U9 are possible suspects, and very common standard LS-TTL chips. Socketing and swapping them may in fact fix the card.

Deunan wrote on 2026-01-24, 09:55:

Other than the two ASICs there is just a handful of 74LS chips. But those do die in weir way, being bipolar based. With good soldering equipment and skills it might be worth socketing and replacing those. The '244 near monitor connector can stay, all it should do is drive the video signal out. The others are potentially suspect.

I suggest to include U2 to the "most likely not guilty" list as well. U2 is a dual 1-out-of-4 multiplexer, which is used to select one out of 4 clocks (only two of them used on a standard EGA), and one out of 4 DIP switches. If U2 fails on the clock half, you will no longer have an image, but as it seems, this card always produces a stable image, even if it fails. If U2 fails on the DIP switch half, the BIOS would have trouble reading the configuration switches, but that part is no longer relevant after POST. A failure of U2 is very unlikely to cause freezes with a stable image during a running system.

When I had the card, that particular pattern of random characters and colours would only appear sometimes during boot, not during games (as far as I remember). The card was working normally, but then for some reason I would do a hard reset, and this would be the first and only image you would see on the screen, the system being unable to proceed any further. Powering off the system and restarting wouldn't necessarily resolve the issue. I have uploaded a high-res picture of this card on the retro web if this helps:

https://theretroweb.com/expansioncards/s/stb- … ems-inc-autoega

I have changed U7 and U8, don't have other two LS's, and i have been played Prince of Persia for almost 45 minutes, and than it froze again. Num lock led is always on, and the whole PC froze (same as before). I need to order 75LS153N and 74LS32N, exchange them and try again. Maybe i will change old sockets also on RAM chips, and solder sockets for another 256KB (have extra 8 chips of 21464), but i don't know which caps i need to solder for this extra 256KB? I can see 50V, but i think the other number is 2 (2uf?). Those caps looks like diodes 😀

mkarcher wrote on 2026-01-24, 11:34:

On the other hand, I have no idea what the empty positions U18..U25 are supposed to be for, as well as the logic chips U7 to U9. The datasheet of the EGA chipset claims that the chip count for an EGA card based on the 82C435/82A436 is just 13, which will be the two core ASICs, the BIOS chip, the monitor output driver chip U3 and the multiplexer chip U2, as well as 8 RAM chips. So the extra logic chips are something STB added to this design for some unknown purpose. Taking an educated guess, it looks like U18-U25 can take another 256KB of video memory for a total of 512KB. also, X1 and X3 could take extra clock generators for modes like 640x480 or 800x600. The EGA chipset used on that card is fast enough to handle those modes if given a sufficiently fast clock. It thus seems your PCB is designed for Super EGA operation, including the option of 512KB, which would provide two pages of video memory even at 800*600, yet only the chips required for standard EGA operation are provided. The three extra logic chips are used for video memory bank selection, and suddenly stuff starts to make some sense. What if memory bus A and memory bus B can be switched individually between the populated standard RAM or the "missing" extension RAM? If memory bus A would be switched to the extension RAM, the screen shot would be perfectly explainable! U8 is a dual flip-flop chip, which can be used to store two bank select bits. U9 is a quadruple OR chip, which can be used to gate four active-low signals (like /RAS or /CAS). This means U8/U9 are perfectly suited to implement bank switching. I can't easily guess what U7 (a 6-bit inverter) might be used for, though.

Most of these Super-EGA cards (and yes, this one is likely influenced by that), came around the time of the DRAM crunch. I'm trying to identify what exact DRAM chips were available at the time... but my thought is generally is the option in layout would have been a half dense chip by 16 chips, giving a total of 256KB still, giving an option if the more dense part is not available. Pretty much at this point all EGA cards were 256KB. I don't recall any that were 512KB, unless they were also a VGA, but this aint a VGA.

Also the strange thing about this, is the late 1991 manufacturing for EGA. It doesn't seem like the high res EGA option was really desired at this point. So this really seemed like a late budget option?? I mean I remember new 286/386 systems still around those days, maybe those did have EGA as an option. But I don't seem to remember anyone buying an EGA card in '91.

The one thing about this discussion I am wondering about is the "auto-sense" feature which when you select EGA in the jumpers, it also supports "multi-frequency" automatically. I am wondering how the auto-sense works, and if a game can trip it up. Is the BIOS dump available? Tracing out how the bank selection works on this PCB will be a clue too if accidental reading is somehow possible.

mkarcher wrote on 2026-01-24, 11:34:

Oh, this picture is actually very interesting. Yesterday I drafted an elaborate response how this might look like video memory a […]
Show full quote

butjer1010 wrote on 2026-01-23, 17:39:

This is the screen when it won't work 😀

Oh, this picture is actually very interesting. Yesterday I drafted an elaborate response how this might look like video memory address bits leaking into video data bits, but the specific pattern you are seeing does not match the EGA architecture at all, but I lost that draft response when inadvertantly closing the browser window, and I re-considered whether this information is relevant to the thread at all, and decided to not re-write it. Today, I read more details of the P82C435/P82A436 data sheet, and actually the pattern very much matches what would happen if the memory data bus A is not driven by the RAM chips during CRTC reads! The P82C435 multiplexes memory addresses and data over the same pins (which is not done at all in the original IBM EGA design), and the character and attribute codes seen in that screenshot exactly match the row addresses driven on the multiplexed pins before the read happens. On the other hand, font reads work perfectly (they happen on memory bus B).

In 80-character text mode, the EGA scan-out logic cards needs every opportunity to access video memory, and fetches a character code and a attribute byte (on bus A) for one character while fetching the pixel data for the previous character (on bus B) all at the same time. This is true for both the original IBM EGA card and the 82c435 integrated EGA controller. As all characters and attributes are wrong, while all font accesses are working, the issue can not be a mismatch between the "high/low CRT bandwidth setting" and the divide-by-2 setting, which could be a symptom if the mode is incompletely programmed into the EGA card. Furthermore, the signal to enable reading of character codes on memory bus A is the same signal as the one used to enable reading font data on memory bus B, it's /OE02 on pin 23 of the 82c435, and as font reads obviously work, the signal /OE02 obviously works, too. I don't see a reason why a card designer would want to add logic that prevents /OE02 to reach character memory (and /OE13 to reach attribute memory), but still allows /OE02 to reach font memory. So maybe the theory of missing /OExx on memory bus A is wrong? Except for /OE, there are two other signals that need to be active for attribute and character memory to respond, which are /RAS and /CAS. If any of /RAS, /CAS or /OE are missing on memory bus A, the screenshot you showed could be explained, but all of these signals are shared between memory busses A and B. So this seems like a dead end.

On the other hand, I have no idea what the empty positions U18..U25 are supposed to be for, as well as the logic chips U7 to U9. The datasheet of the EGA chipset claims that the chip count for an EGA card based on the 82C435/82A436 is just 13, which will be the two core ASICs, the BIOS chip, the monitor output driver chip U3 and the multiplexer chip U2, as well as 8 RAM chips. So the extra logic chips are something STB added to this design for some unknown purpose. Taking an educated guess, it looks like U18-U25 can take another 256KB of video memory for a total of 512KB. also, X1 and X3 could take extra clock generators for modes like 640x480 or 800x600. The EGA chipset used on that card is fast enough to handle those modes if given a sufficiently fast clock. It thus seems your PCB is designed for Super EGA operation, including the option of 512KB, which would provide two pages of video memory even at 800*600, yet only the chips required for standard EGA operation are provided. The three extra logic chips are used for video memory bank selection, and suddenly stuff starts to make some sense. What if memory bus A and memory bus B can be switched individually between the populated standard RAM or the "missing" extension RAM? If memory bus A would be switched to the extension RAM, the screen shot would be perfectly explainable! U8 is a dual flip-flop chip, which can be used to store two bank select bits. U9 is a quadruple OR chip, which can be used to gate four active-low signals (like /RAS or /CAS). This means U8/U9 are perfectly suited to implement bank switching. I can't easily guess what U7 (a 6-bit inverter) might be used for, though.

So, in essence, this means that the broken text mode picture you show would be explainable by a perfectly working EGA card with the bank selection not being properly initialized. It would be easy for the card designer to connect the CLEAR or PRESET pins of the 74LS74 to some RESET signal (e.g. the one on the ISA bus) to make sure the card starts with bus A and B switched to "standard memory", and I would expect this card to be implemented that way. So the "broken text" picture you show might point to a fault in the bank switch logic, and in that case, it is likely not related to Prince of Persia freezing. Well... Are you sure the computer freezes, or does only the image freeze, and you can continue playing blindly? If you can continue blindly, maybe the bank switching logic inadvertantly switched to a mode in which the computer starts updating the extension memory (which isn't there), while the card still displays the standard memory (which will no longer be updated). If that is the case, the "freezing" might also be due to a fault in the bank switching logic!

So indeed, U7 to U9 are possible suspects, and very common standard LS-TTL chips. Socketing and swapping them may in fact fix the card.

Deunan wrote on 2026-01-24, 09:55:

Other than the two ASICs there is just a handful of 74LS chips. But those do die in weir way, being bipolar based. With good soldering equipment and skills it might be worth socketing and replacing those. The '244 near monitor connector can stay, all it should do is drive the video signal out. The others are potentially suspect.

I suggest to include U2 to the "most likely not guilty" list as well. U2 is a dual 1-out-of-4 multiplexer, which is used to select one out of 4 clocks (only two of them used on a standard EGA), and one out of 4 DIP switches. If U2 fails on the clock half, you will no longer have an image, but as it seems, this card always produces a stable image, even if it fails. If U2 fails on the DIP switch half, the BIOS would have trouble reading the configuration switches, but that part is no longer relevant after POST. A failure of U2 is very unlikely to cause freezes with a stable image during a running system.

Yes, my brain doesn't go as deep as quick and I tend to need to look a tonne of stuff up, but it was seeming to me that it's a bus contention, wrong data being fed at the wrong time thing and might be bus release or switching issue. So I would have been looking for latches not latching and that sort of thing. Also might be worth looking for ground floating somewhere giving spurious logic errors, or the converse, high levels being too low or marginal. Could have the caps bleeding too much current off by this age.

By the way: Your card seems to be the "STB systems AutoEGA", and there is a higher-quality picture of that card at https://www.computinghistory.org.uk/det/54777 … aphics-Adapter/ . There also is a high-quality picture at The Retro Web at https://theretroweb.com/expansioncards/s/stb- … ems-inc-autoega provided by the previous owner of the card. The term "Auto" likely refers to the capability of that card to automatically switch to a BIOS-assisted CGA/MDA emulation mode if CGA/MDA software tries to program a CGA/MDA video card using direct port I/O instead of using the video BIOS to do so.

I'm unable to find any resources about a 512KB STB Super EGA ever hitting the market. Also, I was unable to find pictures of the back side of the card, so I am unable to reverse engineer how the bank switching mechanism (if it actually is one) is supposed to work, and whether it might be possible to replace those three TTL chip by a small set of jumper wires to disable the bank switching stuff alltogether. Even without a picture of the back side, I can confirm that the wiring of U18-U25 very much looks like those positions were intended to take another 256K of RAM. I'd enjoy a back side picture of the card to be able to get a better impression of what's going on with U7-U9.

butjer1010 wrote on 2026-01-24, 13:17:

Num lock led is always on, and the whole PC froze (same as before).

Thanks for confirming what you mean be "freeze".I fail to see how anything except the two big ASICs would be capable of freezing the whole PC. When you write "Num Lock LED is always one", do you mean the num lock LED turned on when the system froze, but was off before, or was it on all the time, but you were unable to turn it off after the freeze? In case the latter is true: Did you test whether the Num Lock LED can be controlled using the Num Lock key while Prince of Persia is running before the system freezes? A lot of old games complete take over keyboard handling, and do not allow Num/Caps/Scroll Lock to be toggled while the game is running, so not being able to toggle Num Lock while in a game might not be a sure indicator of a PC freeze.

butjer1010 wrote on 2026-01-24, 13:17:

Maybe i will change old sockets also on RAM chips, and solder sockets for another 256KB (have extra 8 chips of 21464), but i don't know which caps i need to solder for this extra 256KB? I can see 50V, but i think the other number is 2 (2uf?). Those caps looks like diodes

I don't think it makes sense to add extra RAM, because there likely is no software at all that supports the extra RAM. I would not expect the BIOS of your card to support anything beyond classic EGA and auto-switching CGA/MDA emulation. The caps are per-IC stabilization caps on the 5V rail, and the value doesn't actually matter that much. Conventionally you use 100nF multi-layer ceramic caps for that purpose. For the purpose of those caps, internal resistance is way more important than capacity, so do not subsitute them by small electrolytics (like 1µF). They will perform worse.

the3dfxdude wrote on 2026-01-24, 15:18:

Most of these Super-EGA cards (and yes, this one is likely influenced by that), came around the time of the DRAM crunch. I'm trying to identify what exact DRAM chips were available at the time... but my thought is generally is the option in layout would have been a half dense chip by 16 chips, giving a total of 256KB still, giving an option if the more dense part is not available. Pretty much at this point all EGA cards were 256KB. I don't recall any that were 512KB, unless they were also a VGA, but this aint a VGA.

I like the way you are thinking, but except for half-broken chips as used for example for the 32KB "extra memory" in the ZX spectrum, the number of bits in DRAM chips at that time always were a power of four, so there were no half-dense chips, just quarter-dense chips. IBM used 16K*4 chips on the original EGA (in 1984), while the P82C435 only supports 64K*4 chips according to the data sheet.

I know of cheap SVGA cards that use partially broken *4 chips as *3 or *2 chips, and they typically have solder bridges to select which bits of the *4 chips are to be used. On the other hand, this card does not seem to have appropriate signal routing to select bits. Yet, it might be possible that the purpose of the 3 TTL chips is indeed to allow the use of 64K*4 chips as 32K*4 chips, by repurposing one of the address bits as bank select bit. On the other hand, I also think it is possible that the PCB was designed as 512K SuperEGA before it was known that digital monitors lost the competition agains analog monitors, so there was no market for a 512K edition of this EGA card when the card finally was ready to launch. EGA cards were at that time used as low-cost entry-level cards, perhaps combined with EGA monitors from excess stock which were basically obsolete because VGA already won the battle.

the3dfxdude wrote on 2026-01-24, 15:18:

So this really seemed like a late budget option?? I mean I remember new 286/386 systems still around those days, maybe those did have EGA as an option. But I don't seem to remember anyone buying an EGA card in '91.

yeah, that's what I was thinking, too. You don't need a 512K edition of a card that only fits the market as budget option.

the3dfxdude wrote on 2026-01-24, 15:18:

The one thing about this discussion I am wondering about is the "auto-sense" feature which when you select EGA in the jumpers, it also supports "multi-frequency" automatically. I am wondering how the auto-sense works, and if a game can trip it up. Is the BIOS dump available? Tracing out how the bank selection works on this PCB will be a clue too if accidental reading is somehow possible.

As the only crystal oscillator on that card is a 16MHz oscillator, this specific card will not support multi-sync/multi-frequency monitors at all, so I would be very surprised to find BIOS support for any video mode capabilities exceeding plain EGA. The "auto" in the product name is most likely referring to this key selling point of the P82C435/P82A436 chipset mentioned on page 1 in the data sheet: "Smart Auto-Emulation adjusts automatically to display mode required by application software". According to the data sheet, this is implemented by trapping writes to port 3D8 (CGA mode setting) or 3B8 (MDA/Hercules mode setting). These ports are no longer used on EGA/VGA, so any kind of software writing there obviously expects a MDA/CGA card, and enabling BIOS-assisted legacy emulation is a sensible response to writes to that port.

There might have been a niche in the DOS spreadsheet market for high res EGA for a couple of years. Just one more row or one more column on screen was enough to get a win in that field. But there was a budget market for EGA plus discounted monitor vs VGA with high price monitor in 1990ish.

Here is the back side of the card.
If needed, i could make a dump of bios.
So the "whole thing" with num lock is irrelevant , because in POP i can press num lock key, but it will turn on and off, every few times. Sometimes i need to press 3 or 4 times, and then it turns off, sometimes it turns off on first press. Sometimes i can press num lock 50 times and led won't turn off....
Should i consider change those caps near RAM? You said there is no benefit of extra 256KB, so i won't do that

Neither the ground or power pins on that 74ls153 look really well soldered.

BitWrangler wrote on 2026-01-24, 18:08:

Neither the ground or power pins on that 74ls153 look really well soldered.

I removed the LS's, and solder new sockets beneath them. If that was problem before, now it is ok, but still it freezes...

How do other games behave? Commander Keen 4 for example?

keenmaster486 wrote on 2026-01-24, 18:50:

How do other games behave? Commander Keen 4 for example?

Will try right now and let You know

keenmaster486 wrote on 2026-01-24, 18:50:

How do other games behave? Commander Keen 4 for example?

Played for about 15 minutes, no freezes, but screen was teared, like there is no vertical sync at all. Every time i jumped and went to either side, picture was teared. normal walking to side, or jumping straight is ok....

butjer1010 wrote on 2026-01-24, 13:17:

Maybe i will change old sockets also on RAM chips, and solder sockets for another 256KB (have extra 8 chips of 21464), but i don't know which caps i need to solder for this extra 256KB? I can see 50V, but i think the other number is 2 (2uf?). Those caps looks like diodes

I don't think it makes sense to add extra RAM, because there likely is no software at all that supports the extra RAM. I would not expect the BIOS of your card to support anything beyond classic EGA and auto-switching CGA/MDA emulation. The caps are per-IC stabilization caps on the 5V rail, and the value doesn't actually matter that much. Conventionally you use 100nF multi-layer ceramic caps for that purpose. For the purpose of those caps, internal resistance is way more important than capacity, so do not subsitute them by small electrolytics (like 1µF). They will perform worse.
[/quote]

Those 0.33uf i bought for Amiga RAMs aren't ok?

butjer1010 wrote on 2026-01-24, 17:30:

Should i consider change those caps near RAM? You said there is no benefit of extra 256KB, so i won't do that

it seems the3dfxdude was correct:

the3dfxdude wrote on 2026-01-24, 15:18:

but my thought is generally is the option in layout would have been a half dense chip by 16 chips, giving a total of 256KB still

The logic is likely that U8 is meant to latch column address bit 7 (one half latches column address bit 7 of memory bus A, the other half latches column address bit 7 of memory bus B), and U9 is used to distribute the subsequent /CAS pulse to either the first or the second set of 8 memory chips. A complete reverse engineering of that circuit is difficult due to the traces on the top side being obscured by U7-U9, but I've seen enough to confident in claiming that this is indeed not a 512K extension, but a way to use 32K*4 chips - yet I couldn't find any reference that these chips ever existed.

If the card is equipped with standard 64k*4 chips, the whole logic needs to be shut down - most likely by pulling the "/PRESET" pin of U8 low. That pin is driven by the trace with a via inside the silkscreen marking "U14" which then goes around the feature connector into the corner of the card with the serial number sticker and some unpopulated chips. If I guessed it correctly, that trace connects to pin 12 of U4 (a pin of that unpopulated chip). The strange character screen with the correct font look like the upper half of U8 is not correctly permanently preset to Q=1, /Q=0, while the lower half is, which might point to an invalid logic level on that line or a defect in U8. butjer1010 will likely be able to observe whether the "strange character" symptom is gone after changing U8, but even if it is, the freezing issue is obviously not gone.

In case the strange characters re-appear with the new U8 chip, please test the voltage at all pins of U8. Proper operation of that card in "single bank mode" would have

1=high; 2=toggling; 3=toggling, 4=low, 5=high, 6=low, 7 is GND, 8=low, 9=high, 10=low, 11=toggling, 12=toggling, 13=high, 14 is VCC.

So, for the case anyone might be interested, here is what supposedly happens:

U7a inverts /CAS, U7b inverts the output of U7a again. The CAS distribution chip U9 receives the output of U7b. Inverted CAS is used as clock for U8a (trace visible) and U8b (guessed). This makes U8 latch its inputs when /CAS goes low. U7c inverts ground to yield "high" (no idea why they don't just use Vcc), which drives the /RESET pin of U8a (trace visible) and U8b (guessed). That pin (if low) forces the output into Q=low, /Q=high. As this pin is permanently driven high, this feature is not used. i was unable to see whether U7d,e and f are used for anything. U9a distributes /CAS to U22-U25 if U8a is "reset" (input traces visible, output to pin 2 of RN8 guessed; pin 1 of RN8 to /CAS of the unpopulated half of memroy bus A is visible). U9b distributes /CAS to U14-U17 via pins 7 and 8 of RN8 if U8a is "set" (which should be permanently the case). U9c distributes /CAS to U10-U13 B via pins 3 and 4 of RN8 if U8b is "set" and U9d distributes /CAS to U18-U21 via pins 5 and 6 of RN8 if U8b is "reset".

If the memory consists of proper 64K*4 chips at U10..U17, we do not need U8, and U9 can be replaced by a bridge from pin 5 to pin 6, and another bridge from pin 8 to pin 10. If U7d,e,f are not used, U7 can be replaced by a bridge from pin 1 to pin 4 (but only if U9 has been replaced by bridges). This replacement will increase the the margin for memory access time, which is completely pointless and irrelevant unless someone injects a higher memory clock via the feature connector (as the 132-character extension modules do).

butjer1010 wrote on 2026-01-24, 19:27:

Those 0.33uf i bought for Amiga RAMs aren't ok?

They are a perfect electronic fit (0.33 or 0.1µF doesn't matter), and it seems they are also a perfect mechanical fit. As long as you don't intend to modify the card to use 32k*4 memory chips you can not find anyway, there is no need to populate U18..U25 and the respective caps, though.