I did some more digging. It only happens in 256 color mode (even at higher resolutions). Higher color depths don't exhibit the issue. Also don't have this issue in Windows 95A on this machine. So seems driver related and somehow color dithering related at 256 colors?

Tested some more stuff:
- Confirmed my 2.02 VGA BIOS matches copies available online. Same checksum
- Tested all 2mb of vram using X-VESA 1.02 PASSED
- Different waitstates and bus timings. Still have the issue in 256 colors in Win 3.11
- Another driver I found. Still have the issue in 256 colors in Win 3.11
- 86box trying to match this system as closely as possible (ISA SpeedStar 64 2.02 BIOS is available). Does not exhibit the issue with the same drivers I tried on physical hardware

So it has to be some issue with the hardware. But, with the vram testing all good and no other issues using this card outside of the Windows 3.11 256 color issue, I'm at a loss. It appears the corruption is happening during an acceleration function that ends up in the Windows framebuffer since I can capture it directly with a screenshot. But, unsure why it wouldn't show up at 16-bit or 32-bit color modes.

I seem to remember something weird about CL cards and font caching, maybe having to disable it, but details real foggy.

Yes, it would be a good idea to experiment with the "Hardware acceleration" setting...

I've got two copies of this Diamond Speedstar 64 VGA card. It is a fantastic card. Here's mine:

Here is a snoop scan from the first card that has the v1.01 BIOS:

In that scan, a bug about PCI palette snoop register is mentioned. That only relates to use of the Feature Connector port, so that can be safely ignored. (There is a BIOS bug that brings an incompatibility to this card when using it with CRT Terminator, but that can be worked around with a PCI TSR program)

I remember using my card that has the V1.01 BIOS in Windows 3.1 before, and it worked ok.

Now set up my V2.01 card in my 486 PC, and set it up with the CL-GD5434 Windows drivers. I have a file "cirrus_logic_gd5434_(win3.1,_nt_3.1_-_3.5_+_utility).rar" that I've downloaded somewhere (probably from vogonsdrivers?):

Used default settings, except changed monitor brand to "Other brand". I think the only thing that does is it "unlocks" all the options in the other dropdown menus.

Then set 1280x1024@87iHz:

In this test, I was unable to observe the red artifact pixels at least.

Also this card does not provide a boost in Minesweeper skills it looks like.

ahyeadude wrote on 2024-08-11, 19:32:

What's odd is that it isn't corruption on the VGA output, but in Windows? Below are two Alt-Print Screen direct screenshots.

I guess you have memory issues that only appear if memory is accessed at top speed. The ISA bus is too slow for top speed memory demands, so the issue only appears when using the built-in accelerator of the CL-GD5434 chip, that is only in windows. Access patterns of the accelerator may depend on the color depth, which is why you might observe the issue only on 256-color mode. I know there are tools to adjust the accelerator/memory clock on CL-GD542x chips, and you might find a tool for the CL-GD5434 as well. Try if the issue disappears at lower memory clocks. Also, if your card is upgraded to 2MB, you might try without the upgrade memory chips to reduce the load on the memory address/control lines.

Removed the 2 memory chips in the sockets and the issue persists. So it must be the memory soldered to the board.

I removed the two soldered chips with hot air, no issue.

Can I install memory into the sockets with no memory populated on the board? I want to test my chips before soldering them...

Since it does not happen in high or truecolor modes, only in palletted mode, I'd suggest it's not a vram issue, and is likely a pallete register issue.

Soldered on the new memory to the mainboard and still have the same issue.

wierd_w wrote on 2024-08-30, 19:38:

Since it does not happen in high or truecolor modes, only in palletted mode, I'd suggest it's not a vram issue, and is likely a pallete register issue.

The color modes all look fine in VESA modes. No issue with DOS and 256 color games. Even the SVGA driver in Win 3.11 looks fine at 256 color. It's only the accelerated Cirrus/Diamond drivers in Windows 3.11 that causes the corruption. Is the palette register integrated into the chip for this particular board?

I think these are the drivers that I used in my test: https://www.vogonsdrivers.com/getfile.php?fil … 864&menustate=0 . I suppose you gave that driver a try already?

It's hard to imagine that anything else could be the fault than software driver, if all other programs write the palette correctly. But hard to say.

ahyeadude wrote on 2024-08-31, 20:27:

Is the palette register integrated into the chip for this particular board?

Yeah, the palette RAM and RAMDAC are all fully integrated on the Cirrus Logic chip on this graphics card.

This card was made in the 30th week of 1996, A3 revision. Most of the ICs have 96 date codes. Maybe they made some late revision changes that broke compatibility with older OSs? NT 3.51 corruption is similar to Windows 3.11, but worse.

Here's a direct screenshot from NT 3.51. Using the driver from the link clb provided (I've also tried that one in 3.11, same result). I moved the windows left and right a bit, it makes the red corruption worse. Also font corruption is present in NT 3.51.

clb, could you post a dump of that 2.01 bios? I can't find anything but 2.02 out there.

Yeah, hold on a moment..

Diamond SpeedStar 64 Cirrus Logic CL-GD5434 BIOS v1.01

BIOS CRC32: E91A421Fh

Diamond SpeedStar 64 Cirrus Logic CL-GD5434 BIOS v2.01

BIOS CRC32: 8091A10Ch

I double-checked also my Diamond SpeedStar 64 v1.01 card in Windows 3.1, and I do see that it does not get bad palette data with that card either:

Really at a loss to think what would be the cause here.. Got any other motherboard that you could try to install that video card into? Or reinstall Windows 3.1 altogether from scratch?

Thanks clb!

I tried to install it into my Socket 7 motherboard (ASUS P/I-P55T2P4) and it wouldn't post. Not sure what the issue was there.

I've got a hunch, there are two different versions of the board, some with the old Diamond logo on the silkscreen, some with the new. I have the old Diamond silkscreen and an older 5434 chip. But, all the examples online show that with a 2.01 bios versus my 2.02. Mine is likely some sort of transition card. Trying 2.01 is a long shot, but it's something to try.

Ignore the datecodes on the memory, I soldered newer chips on as mentioned previously.