I finally got the new RAMDAC. I took pictures from the motherboard since it must be removed so I can replace the RAMDAC. Here is a picture from the motherboard, here is another and from below.

There was a little surprise below the board. Lots of SMD components and factory fixes.

I started desoldering the RAMDAC with soldering iron and desoldering braid. I carefully removed solder as much as I could but the RAMDAC does not move. Legs are bend under the chip so I don't think I can do much more with this method. I need to use heat gun. I have a good one but it is not meant for electronics and I've never desoldered anything with heat gun. I'll just have to be extra careful.

By the way, someone might notice something odd with that heat sink on top of the CPU. I added that just in case. I had thermal adhesive but it was too old and unusable now so I applied normal thermal paste and glued the sink with three blobs of elastic glue close to corners. This way the heat sink can be easily removed later. I've never done that before but it seems to work just fine. Also, the CPU is 486DX-33 but not the one that came with the computer. I had IBM (manufactured?) 486DX-33 (with Intel logo) that had smaller cover under the CPU (like the one on DX2 CPUs). So the IBM one is manufactured with smaller process. And yes, this IBM one actually runs cooler but I still applied heat sink on it. Also the motherboard had EPROM BIOS chip but I replaced it with FLASH chip when I put XTIDE Universal BIOS in it (I want to preserve the original).

New RAMDAC succesfully installed. And I also put the 25/33 MHz switch to the back of the computer. Unfortunately I first made a hole too high so I had to make a second hole. Here is one picture and here is another.

The new RAMDAC is Bt 478KPJ80. It is actually more capable that the UMC UM70C171L-65 it had originally. Bt is 80 MHz RAMDAC and the UMC is 65 MHz. It should not matter here because faster means higher resolutions and refresh rates so not any use for DOS or games.

The new Bt is actually 8-bit (RGB888=24 bit) RAMDAC while the old is 6-bit. But... Before ordering the chip I had to make sure it is compatible. Datasheet tells the Bt 478 is 471 compatible but there is one pin that must be connected to ground (it selects between 6- and 8-bit operation mode). But when I removed the old RAMDAC there was 471/478 printed to the PCB. And indeed, it was wired so that 478 is set to 6-bit mode if one is used. So I did not have to make any modifications at all.

Removing the old RAMDAC was not that easy. I had to use the heat gun and it was just too big. I set temperature to 380 C and blow to minimum but it was just too much. I used tin foil to cover nearby components but I melted one corner of printer port just a little bit (and it was protected by tin foil). At that point I stopped using the heat gun. One edge of the chip was still soldered but I was able to desolder it eventually. I almost destroyed three soldering pads but I guess I was lucky that there was not any permanent damage.

Next I soldered a socket as I didn't know if the new RAMDAC is working or even genuine. Soldering a socket was easy compared to removing the old RAMDAC but it took some time to get it soldered.

I first tested with the old RAMDAC and it still worked! I feared I had destroyed it but it survived. Then I installed the Bt and it also worked! And the jailbars were gone! Just one thing to note. The old RAMDAC has pin 1 above the text while the new has pin 1 at the left side of the text. So it is very easy to install it the wrong way.

Finally it is time to install Hard Disk and CF reader. I say finally because this took a lot more time than I thought it would.

I made 4 of those but only one is ready. The rest needs to be assembled. I painted them with this (since I happened to have it already) but the color is not correct. So I added drops of black to made it darker. With black added the color way too gray or cold. So I added drops of orange to make it warmer. I think I painted them seven times and 5 times I adjusted the color a bit. Well, many layers of paint make it look more plastic. In fact now when one is installed it does not look wooden at all and that was what I was after.

But the texture is a bit off. It should have been much more smooth. I intentionally tried to avoid it to be too smooth but it is too coarse now. I should have used spay gun but instead I pained by tapping with a small sponge, one that is used to wash dishes. It just made the texture too coarse. I sanded a little after each layer of paint. Even after the last layer to make the texture smoother.

When I finally got one assembled, I noticed the 1,8" HDD adapter and CF adapter have pin 1 on opposite sides. So I need to twist the IDE cable and that requires more space. It took some time to get it installed but I think the result is great! The 1,8" HDD works as master and CF as a slave drive. Of course some CF cards and even microdrives do not work as slave drives but the excellent Hitachi microdrive does so I'm happy. For some other computer I could have used two cables. One to connect the 1,8" HDD to primary IDE and CF adapter to secondary IDE.

Here are more pictures of the drive.

That's a really neat solution - I like the combined 2.5" drive and CF reader in the one bay 'module'. Well done!

Actually it is 1,8" drive but I'm sure it would have been possible to use 2,5" drive and adapter as well. I wanted to make drives like this since it is very hard to access CF cards if they are behind computer. Also cables are usually not long enough to connect HD as master and CF slot as slave if it is at the back of computer. So secondary IDE would be needed. And that is likely used by CD-ROM and again the cable is too short so the CF adapter at the back of the computer would need IDE controller of its own. XTCF card could be used but it is connected to 8-bit ISA and is slow. So the idea was to include CF-adapter and HDD as a single drive that would minimize the cables.

I could have bought 3,5" bay CF to IDE adapters and modify them but they seem to look awful and are expensive. Certainly my self painted are closer to what they should be. If I'm going to make more of them they'll be even better.

I use the Startech 3.5" bay adapters myself, but I always paint them to match my cases:

They're only available in black these days, so don't really match any 'usual' cases of the period.

Those look really nice. Do you happen to have proper color codes for the paint? I'd like to try repainting a case or even monitor someday and I'd like the color to be what was originally used.

I got rest of my CF-HDD-drives ready and I turned the CF adapter around so now the cable goes nicely. Here are couple of pictures. I'm not completely happy with this either. I should have turned the 1,8" HDD adapter instead. It would have hidden the HDD label but the IDE cable would have fit inside my CF-HDD-drive. Also I think I should have used those CF adapters with bracket (without the bracket of course). It would have required more modifications but the PCB would have been easier to mount properly. Maybe rev 3 will be near perfect then but these will do for now.

I'm still testing the Copam with case open. I did install PC-DOS 2000 and games and I'm now doing some testing. One thing I did notice is that the HDD seems to automatically enter some sleep mode. It is not unexpected that it has such feature. XTIDE Universal BIOS has power management support but it is for enabling it. I'll need to check if it is actually left at drive default when it is set to disabled from xtidecfg.com.

I was fortunate to actually paint the entirety of that second case when I bought it, so I know the exact colour 😀

With the first, I had to try a few shades until I found a decent math - it's not bad as long as you look at the front, but the right edge, and to a lesser extent the top of the bezel is yellowed.

I had to put the UMC RAMDAC back last night for testing. I noticed some graphics errors (color dots) with the Bt RAMDAC and I was pretty sure the UMC did not have them. And it didn't. These dots that the Bt displays are usually visible while loading etc and not during gameplay. So these are related to palette changes. I have noticed them on some graphics card before so not a big deal. Many games do not have any issues but I found couple of really bad cases. Star Control 2 has completely wrong colors but there is the /g:bios command line switch that fixes them. What it does (I believe) is it uses video BIOS to write to palette registers instead of directly writing to them. But the other game, Monkey Island 2, does not have such command line switch. I don't remember using a computer before that would display Monkey Island 2 wrong.

So I retested with UMC RAMDAC and everything was just fine (except it has the jailbar issue). I cleaned the Bt RAMDAC properly (and printing did not come off so it is very likely to be authentic chip and the printing did not look new to begin with) but cleaning the contacts did nothing at all. As I mentioned I have seen similar issues with some graphics cards before but not this bad. Since the the HT216 is integrated to VLB it might be possible that it is too fast for the Bt RAMDAC. Maybe it would work better on a ISA card?

I got Digital DECpc 466D2LP recently. I had to test it a bit to make sure it works. It had broken floppy drive that destroyed disks and two Quantum 240MB drives that both worked on first testing day but the other did not next day. I was able to run Phils 386 benchmarks pack on it after I replaced the drive. It revealed that the onboard S3 924 was integrated to ISA and not VLB as I expected. But this computer happens to have a RAMDAC I could try on a Copam and I could also test if the Bt actually works better on ISA. But the RAMDAC is soldered and it was already hard enough to desolder the UMC from Copam. Another alternative I have is ATI VGA Stereo F/X but I definitely don't want to experiment with it.

This is from Bt datasheet: "Occasional accesses to the color palette RAM can be made without noticeable disturbance on the display screen; however, operations requiring frequent access to the color palette (such as block fills of the color palette) should take place during the blanking interval."

And this if from Sierra Semiconductor (ATI has one of those) datasheet: "To reduce noticeable sparkling on the CRT screen during MPU access to the color lookup table or the overlay registers, an internal anti-sparkle logic is implemented to maintain the previous output color data on the three D/A Converters output while the transfer between the color look-up table RAMs and the RGB registers occurs."

So I guess I'll need to try and find Sierra RAMDAC next.

I took a look what the RAMDAC in the DECpc is. It happens to be Sierra semiconductor RAMDAC but too good one, SC15025. It doesn't even support 6-bit mode so I can forget that.

I did more testing about possible speed sensitivity of the Bt RAMDAC. Setting CPU to 16 MHz did not help but disabling internal cache did help, even when CPU was set to 33 MHz. Monkey island 2 colors are all correct when internal cache is disabled. So I'm even more confident that the Bt would work much better on a ISA card. I'm leaving the Bt in Copam for now. UMC would be better if I'm going to use the old 12" IBM VGA monitor but I haven't decided yet. I'm hoping to find one of those suitable Sierra RAMDACs but who knows when that happens.

New RAMDAC is now on its way. Actually I found a whole ISA card based on WD90C30-LR controller. It is very good, actually. And not only that, this card has RAMDAC already on a socket so it makes things a lot easier.

What makes the WD90C30 so good? It is fast and has good compatibility but most importantly, I already have one card based on that controller and it is the fastest card I've found for my 286. Yes, the one I have has 286 compatible BIOS. Most better ISA graphics cards have a BIOS that uses 386 instructions to rule out 286 systems. Second best VGA card I've tested on my 286 is the Ati VGA Stereo F/X, but it has worse VGA compatibility and it is slower.

That is not all, the card I have has Samsung RAMDAC but on a DIP package so I cannot test it on Copam. The new card has the very same Samsung RAMDAC but on a PLCC package that fits in Copam. I can also easily test if the Bt actually works better on a ISA card. I took a look at the Samsung datasheet and it should be snow free (and I don't remember seeing any issues on my 286) but if for some reason the Samsung RAMDAC does not work on a Copam, then at least I have and excellent spare card for my 286. I'm going to test the card on a Copam as well since I'm interested to see how it compares to the S3 928 I already tested. I'm going to test Cirrus GD5426 as well for even more comparison.

I was actually testing something else when those Bt issues came too noticable. I was testing if General Midi card is needed. Since this system is for games before Doom and CD-ROM-era I was under the impression that there are not many General Midi games but a lot of MT-32 games. That was the case. If General Midi was supported, MT-32 was also. And many games expected the General Midi device to use port 330 and that is the same where MT-32 should go. So even though Orpheus allows to use two ports, one for intelligent mode and the other for non intelligent mode (enough for General Midi devices) I would have to use the same port for both to maximize General Midi support to make the General Midi device actually useful.

Using same port for both does not matter that much. I have Roland SCC-1A on my 286 as a controller for old type MT-32. By default sound comes out from both but there is a sysex to silent the SCC-1A so only MT-32 will produce sound. And if I want sound only from SCC-1A, then I simply do not power on the MT-32. Quite simple but it would have been even simpler if two different ports could be used. Later games usually always allow to configure the midi port so for late 486 or Socket 7 systems I would have put the General Midi device to port 300 and used MT-32 with all games that require port 330.

Speaking of sysex, I wonder if it is possible to make a sysex that would only play the extra CM-32L instruments. That way the MT-32 would play whatever it plays but the GM device would play the missing instruments if game supports those. That would make the General Midi device much more useful and it would make a lot of sense to set them to same port.

But anyway, the conclusion is that I won't install any General Midi device to the Copam. It can have the old type MT-32 since I finally found LAPC-I for my 286. So the Copam will have Orpheus, GUS Classic and MT-32. What a perfect setup for a computer meant to play games before Doom and CD-ROM-era. By the way, GUS can act as a General Midi device if it turns out that some game needs it after all. I think I should write more about GUS next.

aitotat wrote on 2021-07-03, 05:46:

..
fastest card I've found for my 286. Yes, the one I have has 286 compatible BIOS. Most better ISA graphics cards have a BIOS that uses 386 instructions to rule out 286 systems. Second best VGA card I've tested on my 286 is the Ati VGA Stereo F/X, but it has worse VGA compatibility and it is slower.
...

I've never seen this before. I have several 286 boards and have used all kinds of VGA cards in them - rubbish Oak things, Tseng Labs ET4000AX, and various late model Cirrus Logic cards (5428 and 5429) between which are by far the fastest ISA VGA devices.

None of them refused to work in any of the 286 boards.

I must have had back luck then. Slow card have worked but 1993 cards have not. I've tested one CL-GD5422 based card and it did not work but worked on a 386 and 486. The #9 S3 928 I already tested on Copam did not work on my 286. Then there was one Ati 28800-5 based ICL card but again, it did not work on my 286 but worked on 386 and 486 computers. I was surprised that the 28800-6 based Ati VGA stereo FX did work since the ICL did not. I also tested Trident 9000 based card but I don't remember did it work or was just too slow. Likely it did not work but it did work on later systems. I had to use WD PVGA1A based card before I got that ATi VGA Stereo FX. Actually PVGA1A is a good period correct card for a 286 so it was not that bad at all. But it is no match for later cards.

I also tested ET4000-based card. It did work but had graphic issues that were not present on 386 and 486 systems. Now that I think about it, it is possible that it had the same RAMDAC issues that I'm now having with Bt. One thing I like about 286 is super fast ISA speed that is not reachable on later systems. Maybe the ET4000 had Bt as well and the speed was too much (I have 16 MHz 286 on a HT12 based motherboard with zero wait state RAM). Well now I definitely have to test the WD90C30 with Bt RAMDAC on a 286 as well.

I'm surprised to hear that even CL-GD5429 can work on a 286 since my 22 did not. I have EPROM programmer now so I can test other BIOSes to find a compatible one. I just recently got the CL-GD5426 based card so I'll start to experiment with that. I no longer have the other cards I mentioned, except the S3.

I updated Copam benchmark results with CL-GD5426. It is faster than S3 928 and almost as fast as VLB based integrated HT216.

I was last of the early birds to get McCake. It arrived yesterday. Fully MT-32 compatible MIDI daughter board is quite unbelievable. And it supports soundfonts as well. I didn't have time to test it yet but I'll try to do so today. It should be an excellent board for any MS-DOS gaming system, Copam included. I don't really need it for Copam since I have the external MT-32 for it. McCake likely goes to my Super Socket 7 system. It has 550 MHz K6-2+ and Voodoo 3 so it can run early Windows 98 games nicely, even though it is meant to handle end of MS-DOS era (everything that my 100 MHz DX4 system cannot do). I have Yamaha DB50XG on the SS7 machine at the moment. The Yamaha is there in case I want to play Final Fantasy 7 with XG MIDI support. But there are XG soundfonts that McCake can use so maybe the McCake could be of best use in the SS7 system. I'll just have to try.

But what if I wanted to put the McCake in Copam? There are some problems. First I would not want to try fit it between Orpheus and GUS Classic. And GUS Classic does not fit to the other ISA slot without extra trouble. And if Orpheus was on top and McCake on top of that, then there would be the problem that I wouldn't be able to see if weight of the monitor somehow presses McCake. But I could use a cable and relocate the McCake elsewhere. Only place would be where 3,5" hard drive used to be and I wouldn't even need a long cable for that.

Next there is the heat issue. After all McCake is a computer of its own and ventilation is already quite minimal. Perhaps best solution would be to make a hole under the 3,5" hard drive tray and put a fan there. It would directly cool the McCake, and also the CPU would be near too. And I wouldn't have to do any visible holes anywhere. Better ventilation would be a good thing even without McCake. The problem would be how to actually make a large hole. This computer is heavier than it looks so making a large hole would require good equipment.

After all that there would still be a problem left. PSU is only 65 watts and I don't know how much is for 5V and how much for 12V. It is likely that McCake would be too much and I'd have to replace PSU. So McCake would require too much modifications for this machine.

New RAMDAC arrived, or actually 16-bit Tandy VGA card with WD90C30-LR controller. It is no wonder similar card performs so well on my 286 because it is fast, just as fast as CL-GD5426. I didn't expect it to be that fast. I updated results to the spreadsheet. WD controller is older, maybe one year or so. Anyone looking period correct cards should definitely check out the WD90C30-LR since it performs so well. And it has perfect VGA compatibility as well, just like the Cirrus (and unlike the integrated HT216).

The Samsung RAMDAC is KDA0476PL-80. I first tested if there was any of the issues when it was on the ISA card and I was a bit disappointed. Based on the datasheet it should not have any artifacts but it did have a little, not much and nothing significant. It is way better than Bt but not as good as the UMC.

But it also has color palette issues with Monkey Island 2 and Star Control 2. Again both can be fixed by disabling internal cache. I also tried to set CPU clock to 25 MHz but it did not help. The color issues are far less severe than the Bt had but you'll definitely notice if even one color is wrong. Fortunately it doesn't seem to affect many games but more testing is needed.

Next I tested the Bt RAMDAC on the Tandy WD90C30-LR card. The color issues are lesser than with the integrated HT216 so ISA bus does help there. I'd say the color issues reduced to Samsung RAMDAC level. But the Bt artifacts are present and just as bad as before.

How about the Samsung RAMDAC on the integrated HT216? Just the same as on the ISA card, not any better and not any worse. Samsung RAMDAC is a compromise between the UMC and Bt. It does not have jailbars but it does have some of the issues. Artifacts are minimal but I don't like the color issues one bit. I think the Bt does have sharper image but the Samsung has good enough image quality.

I'll keep the Samsung for now and I don't actively try to find yet another RAMDAC but I'll test one if I happen to get one. And I hope to eventually find one without compromises (if someone does have good RAMDACs to sell, please send me a PM).

I did a little more testing last night. Color problems are gone when CPU is set to 16 MHz but they appear again at 20 MHz. I could modify the switch at the back of the computer to 16 / 33 MHz (change of two jumpers is needed instead of one). 16 MHz is faster than 33 MHz with internal cache disabled. Another temporary fix should be possible with software. It should be possible to use EMM386 port trapping to trap RAMDAC ports and add a small delay. Well, I don't think any delay is necessary after the API overhead. For the Bt it could be used to write the RAMDAC registers only at blanking intervals.

But the EMM386 would need to be loaded and I don't like to do so unless necessary (but it might be necessary anyway since there is no UMB driver for this Headland chipset). TSR would steal some RAM but with EMM386 loaded that would not matter anyway and I could even place the resident code to ROM. But the EMM386 trick would not work with protected mode software. Those problematic games at the moment are real mode games so perhaps the TSR would be better alternative than to slow the computer to 16 MHz.

Of course the real fix would be to find RAMDAC with RAM part like UMC and DAC like Bt. But perhaps the TSR would help some other as well. I know I'm not the only one with the problem. However many likely do not know what the problem actually is or how to fix it.

I'm actually trying to code such TSR but no luck so far... I can trap the ports and non-problematic games continue to work fine. But the problematic games now have even more wrong colors even when internal cache is disabled! Results are the same even when I don't do anything in the EMM386 handler excepts return with CF set so default handler does the I/O.

I have no idea what those games do differently or why trapping ports 3C7h-3C9h makes things worse.

I got the port trapping somewhat fixed. Meaning colors are better like when no trapping at all, so colors are not quite correct on the problematic games. I think the problem was that there is some time frame to output all red, green an blue values and that was not met when trapping ports and delaying access to them. The fix was that I buffered all three color components and output (and input when reading) all of them at once.

But that was not the fix I hoped for. I tried a lot of things, like waiting for retrace , but no success. I swapped RAMDACs few times between both the Copam and the WD90C30 based ISA card (BTW the UMC was just as bad on the WD). My port trapping was not any better on ISA card even though ISA should not be too fast. I even tried to remove RAMDAC completely from Copam to make sure it was not interfering with ISA card but no change.

Then I remembered the unexplained Sound Blaster Pro 1 and 2 mixer issues and decided to try the WD card with Bt RAMDAC on the "side project" Cyrix DLC-40 machine. And no color problems at all! Sure, the Bt had the snow as expected but otherwise it was just fine.

So the problem must be with Headland chipset and it affects both ISA cards and the VLB integrated graphics card. I wasted quite a lot of time. Well it was not completely wasted since now I have some experience with port trapping.

Honestly, the problem is very minimal. Few games need to be slowed with the Samsung RAMDAC and everything will be fine. Still I'd like to actually fix the problem, what ever it is. Time to read Headland datasheet. Maybe some wait state will fix things but I'm not going to spend days with this, thats for sure. I'd finally like to play some games.

Headland datasheet did not help. I tried a few things with DOS debug but nothing helped. So I think hardware is ready for now. I'll try another RAMDAC if I find one. But for now I think this computer is finally ready. But one more thing about the RAMDAC issue. I should have tested again with the DX2 processor underclocked to 33 MHz. It would have been interesting to see if 16 MHz bus alone would have fixed the issues as underclocking the DX CPU to 16 did. But I already closed the computer and moved it to more suitable location before thinking about it.

So the computer is ready for now. I changed the picture on the first post but here is a link to it. I moved the computer to where my 286 used to be. Notice the MT-32 (old type 1.07) on top and a new monitor! I found a very nice 15" Belinea 10 50 35 for only 5 euros. It is very hard to find 15" and smaller CRT monitors here in Finland. If you find a CRT monitor, it is likely one of the cheap 17" models. In my opinion 17" is the worst possible monitor size for DOS gaming. It has scanlines that were not supposed to be visible on VGA because of scanline doubling (but CGA and EGA monitors should have very visible scanlines so I suppose those modes are displayed wrong on a small VGA CRT). If scanlines do not matter, then 19" is better size (if you can find a good one without geometry issues). If you don't want scanlines, then 15" is the maximum size. So nothing good in 17", unless you really want scanlines and 19" is just too large for your table (and they can be).

While I don't like 17", I just had to get one when I found new old stock Viewsonic E70 couple of years ago. It is not in use at the moment but I'll try it on my DX4 system again, especially now that Copam is equipped with 15" CRT.

But back to that 15" Belinea. It is manufactured in 1998 but external design looks a lot older and it looks just fine with Copam, well maybe 14" would have been a better fit but this is not too large. Belinea had surprisingly sharp picture but color temperature was colder than usual and OSD had no color temperature settings at all. When CRTs were the norm I used the cold (9300K if I remember correctly) color temperature that was the default with CRTs. It gave the sharpest (most readable) picture but now I want to use warm temperature (6500K I believe it is) for games. Games just look better with warm temperature. It may look odd a little while but you'll get used to it very soon and don't want to go back. After that 9300K looks almost like colors are missing. Anyway, here is picture from my 19" Dell monitor with flat Trinitron CRT. So lovely colors (also notice the scanlines).

Since flyback transformer needs to be usually adjusted on CRTs these days (I won't write anything else about it. It is the dangerous part you should not touch!) , I'll decided to try adjusting the colors while opening the monitor. I hoped to get it to something like on the Dell. First I adjusted just red and closed the monitor. Later I noticed that black had a very little red tint on it when room was dark. My first mistake was that I did the adjustments on a daylight. It should have been made on a dark room.

So I opened the monitor again. I happened to have CGA compatibility tester on hard drive and nothing more suitable for adjusting. But it was actually very good for that since it displayed all CGA colors and had text what they should be. It was actually quite hard to get brown to look brown (it was easily almost red) and dark gray to look dark gray without a tint on it. Actually it now has a very little red tint. I did want warm colors after all. So eventually I had to adjust not only red but green and blue as well. This is the result. If you direclty compare it to the picture from Dell, it does look like it is a bit too red. But it looks fine if you don't have anything to compare it with. So games look just fine now! Definitely a huge improvement to what it was (unfortunately I did not take a picture at the beginning but here is a picture after first red only adjustment).

This was the first time I did CRT color adjustment and I'm sure the result is not correct like when done with monitor calibrator.