I was hoping this strange pattern of characters would ring a bell with someone here - I can´t believe this is such a rare graphics defect...
However, here are some more details:
The graphics chip is a S3 P86C928.
There´s also a version P86C928P, that´s supposed to be the PCI version, but I found it on ISA and VLB cards also. Would this be a pin-compatible replacement? I´m asking because the P-version is still available. There also seems to be a version with just "P86C928P" and another one with an additional "PCI" marking besides the "S2" logo.
Could a faulty RAMDAC also produce this pattern? Or a defective GAL?
I think I can rule out interrupted traces and bad solderings.

Is the video ROM shadowed? If it is, does turning it off change anything?

It certainly looks like a corrupt font or similar (I had similar text display corruptions with a C&T 65550 which apparently is missing a bios font), which is why speedsys is fine - if I'm correct speedsys is likely not using bios fonts.

Any problems at all when running graphical stuff?

majestyk wrote on 2024-04-13, 17:04:

I was hoping this strange pattern of characters would ring a bell with someone here - I can´t believe this is such a rare graphics defect...

similar Re: "Fixed" 386sx motherboard works but not with 16-bit VGA card while using modern SSD/CF on old ISA system (ALE pin)

I think the linked issue was mainboard related, I had tested my ELSA card in different mainboards both ISA and EISA - always the same problem.
Besides that I found the graphics chip to become quite hot in intervals for a few seconds - depending on whatever.
Suspecting a broken chip I searched for a while, but the only way to obtain a replacement P86C928 was buying another graphics card which is pointless due to the sky-high prices.
So I ordered some new 86C928-P chips out of the blue. There seem to be two different versions, one with additional "PCI" marking and the second with just "-P".
Today the chips arrived - not the PCI version luckily. (Usually you can read that -P is the PCI version, but the -P version without "PCI" printing can be found on ISA, VLB and EISA cards.)
After replacing the chip the card is working perfectly now in both ISA and EISA mainboards.

This is probably the only ELSA XHR Winner 1000 with a "-P" chip version 😀

majestyk wrote on 2024-04-11, 06:43:

There seem to be certain rules at work, like when I type numbers: 0->0, 1->0, 2->0, 3->3, 4->4, 5->4, 6->4, 7->7, 8->8, 9->8.....

Check for a short between video memory data bit 0 and 1. All the replacement patterns you quote have these two bits at the same level. As the '928 is a VRAM-based system, the issue might be in the serial read-out path or the parallel memory access path. As you don't seem to get any diagnostic memory errors, I suspect the serial path. All bus-based video memory reads and writes are performed through the parallel path, only displaying video can be performed using the serial path. The '928 is in a 208-pin PLCC package. If you look at the chip with the letters in "normal orientation", pin 105 is at the right end of the upper edge and pin 156 is at the left end of the upper edge. Serial data pins 0 and 1 are at pins 145 (SID0) and 144 (SID1).

In case the parallel video memory interface is causing the trouble, the relevant pins are on the left edge of the chip. The pins on that edge are numbered from 157 (top end) to 208 (bottom end), with PD0 at pin 183 and PD1 at pin 182. It might be helpful for orientation that pin 184 (below PD0 and PD1) is grounded. The same is true for pin 146 (left of SID0 and SID1).

majestyk wrote on 2024-04-19, 18:35:

Today the chips arrived - not the PCI version luckily. (Usually you can read that -P is the PCI version, but the -P version without "PCI" printing can be found on ISA, VLB and EISA cards.)

The "PCI" just indicates that the chip support the PCI bus if it is strapped for PCI support. The datasheet for the non-P 928 I have at hand only defines bus type 0=EISA, 1=386/486 local bus, 3=ISA, but it leaves out 2, which has later been defined to be PCI. So as long as a card does not strap the bus configuration pins to mode 2, which is reserved on the non-P, it will not run the -P version in PCI mode.

Thanks! Finally the different versions are understandable. Note that the date code of my P-chips is 1995 (as opposed to the original from 1993) - PCI had been introduced in the meantime.
One of the 3 245 bus transceivers ( the right one) had to be replaced. It must have exploded and it had a hole at the top. That´s why this one is now "ALS" the others are "LS".
I´m sure the card has been mishandled in the past and at the beginning I hoped the defective 245 was all that´s broken, but - of course - it wasn´t.