Finally figured out whats is the reason of brightness issue on S3 cards: the BLANK pedestal.
According by Savage4 datasheet, it can be toggled by bit 3 of SR27:

This works on Trio3D famly also, for ViRGE/DX and Trio64V2/DX it can be toggled by bit 5 of SR1A.

So, for Trio3D and higher, the DEBUG.EXE command sequence is:

1o 3c4 08
2o 3c5 06
3o 3c4 27
4i 3c5
5<at this step subtract 08h from given value>
6o 3c5 <resulting_value>
7q

Probably BLANK pedestal can be disabled via BIOS modding, will check it later.

Hmm I should take a look at the noise and dynamic range with and without gain.

I think that with PCs and monitors, the pedestal should be off. It would actually be nice to compare video signals with oscilloscope.
Normally the term pedestal is familiar from NTSC video, but it's possible it can be used to have room for sync-on-green signals as well.

Most likely the register is loaded from a table containing sets of register values - at least the standard registers are usually loaded like this when video mode is changed. If there is a bios update available with too bright colors, it should be easy to see what goes wrong.

I checked with an S3 Trio3D/2X and the Epiphan VGA2PCIe. With optimal test image dynamics I get with reg 27
set to 00: Offset: 32/32/32 Gain: 177/176/179 Signal rise time 6.98 ns / 143 MHz
set to 08: Offset: 34/34/34 Gain: 190/188/192 Signal rise time 7.9 ns / 126 MHz
Reference with a Matrox Millenium: Offset: 32/32/32 Gain: 185/190/188

Offset is a measure of the voltage of 'black', Gain Values of 255 would be minimal gain and 0 is maximum Gain of the capture card.
So the black point shifts indeed where the image appears brighter on the same settings.
It seems also that the signal is sharper with a setting of 00.

patched video BIOSes for S3 Trio3D/2X with BLANK pedestal disabled (in attachment). Tested on PCI card with 4MB of SDRAM, should work with others. MCLK for SDRAM\SGRAM = 90 MHz

Could you compile a small tool as well that allows to set the bit 'ON' and 'OFF' from plain DOS on the fly?

elianda wrote:

Could you compile a small tool as well that allows to set the bit 'ON' and 'OFF' from plain DOS on the fly?

Here are two simple DOS .com programs to either turn the bit on or off, but I haven't tested these. I used NASM to compile these so I haven't tested them on S3 card. These are so simple you could just do these in debug and save them.

Jepael wrote:

elianda wrote:

Could you compile a small tool as well that allows to set the bit 'ON' and 'OFF' from plain DOS on the fly?

Here are two simple DOS .com programs to either turn the bit on or off, but I haven't tested these. I used NASM to compile these so I haven't tested them on S3 card. These are so simple you could just do these in debug and save them.

these utils works wonders on the Diamond Speedstar A55 AGP card , thanks for that!

But is there a way to patch the BIOS and make this permanent? I can post a dump of the BIOS easily, can anyone do it so the Diamond BIOS/Branding is intact? 😊
This one is still bright even after entering windows... 😵

The cards befor Virge seems have no pedestal setting. I'm not sure they had no similar image issues, as developer understood what he does and hence such image was intentional.

I am attaching the A55 Bios, version 2.0C.10
Did some compare between this and the patched BIOS wbc posted some posts above, from what I could see the real difference (except branding) is 5 bytes at offset 0x119A (original is 74-04-80-CC-08 and the modified bios is 90-90-82-E4-00) , is that the only change?
There is also a single byte at 0x7FFF (10 in original vs 61 in modified) so I guess that's a checksum? [edit: must be, changing only the 5bytes at 0x119A did not work at all, the screen was blank but the computer booted just fine)

Help is welcome 😊

I got two cards with this issue.

One has a socketed chip, the other one is soldered. I have one card that doesn't have the issue. Might have to dump the BIOS of these cards at some point.

keropi wrote:

I am attaching the A55 Bios, version 2.0C.10
Did some compare between this and the patched BIOS wbc posted some posts above, from what I could see the real difference (except branding) is 5 bytes at offset 0x119A (original is 74-04-80-CC-08 and the modified bios is 90-90-82-E4-00) , is that the only change?

AFAIK yes (there is one byte at offset 0xD4 == 0x02 in original image and 0x00 in patched but it is useless by fact).

keropi wrote:

There is also a single byte at 0x7FFF (10 in original vs 61 in modified) so I guess that's a checksum? [edit: must be, changing only the 5bytes at 0x119A did not work at all, the screen was blank but the computer booted just fine)
Help is welcome 😊

0x61 is a checksum for 362_C10E.BIN which of course differs from A55 video BIOS (even if patched), so checksum MUST be recalculated after patching.

Anyway, try this BIOS image 😀

^ thanks for the epxlanation wbc - the patched BIOS works fine so thanks a bunch for this too! 😊
Did some benchmarks with Phil's DOS suite, in my 233mmx system the A55 scores the same as the Riva 128zx (so bottleneck is the rest of the system) so it's a pretty decent vga now with the patched BIOS.

Just in case anyone wonders, the original bios chip is a Holtek OTP HT27C512/70ns - so no need to remove the label. Just use another 27C512-style (e)eprom and it work just fine

Been testing all my S3 cards (90% of them are Diamond ones) and I found 2 that are problematic:
a. Powercolor VirgeDX/375 4MB, this works fine with a bios from a Stealth 3D 2000
b. AGP Trio3D/2X 86C368 generic vga, I dumped the bios since it specific mentions 86C368 , maybe it's different from the 362 one? Version is 2.0c.0983. Tired the patched A55 bios and it works but maybe the generic 368 one is of interest?

It would be nice if some additional information about these BIOS images would be available. Like the clock rates for Chip/Memory the BIOS sets.

Keropi, did you dump the BIOS with a flasher or via software?

^ always using a programmer for this kind of work 😉

keropi wrote:

b. AGP Trio3D/2X 86C368 generic vga, I dumped the bios since it specific mentions 86C368 , maybe it's different from the 362 one? Version is 2.0c.0983. Tired the patched A55 bios and it works but maybe the generic 368 one is of interest?

erm...

So 86C362 and 86C368 are fully compatible I assume (since I'm also using 86C362 video BIOS on 86C368 based card 😀)

P.S. by the way I'm finishing my TSR fix utility (i.e. fixes some Quake bugs and other bugfixes for ViRGE/DX and later), so coming soon 😀

heheh 😁 I did not bother too look further down the bios image, it does state 368 in the beginning 🤣
by Quake bugs you mean the stripes on the right side? I noticed them today with some resolutions, I am making the switch from riva128zx to A55 on my ss7 build (3D is of no importance to this build for me)

keropi wrote:

by Quake bugs you mean the stripes on the right side?

if they appear in 360-wide modes then yes. I've also got it on ViRGE/DX if you set 320x200 VESA mode then any 360-wide mode.

UPD: and that TSR is released now!