Apologies for necroing an old thread but I've been playing around with an original S3 ViRGE and I asked ChatGPT whether it could write a small C program that automatically checks and corrects invalid checksums for ROM files. After a little back and forth it came up with the attached which worked for me- much quicker than doing it manually. Included is the C file as well as a macOS (Intel) and a Win32 console versions.

Very useful thank you!

Sorry for the bump.
I have revised interest on s3 older cards.
So far what we know is this:

Bios 1.01.03 has the brightness bug at least the version I have and it supports just VBE 1.2.

Bios 2.01.07 hasn't brightness bug and it supports VBE 2.0
Bios layout and mclkconv from wbc: Re: How do I flash VBIOS on an S3 card?
One more thing, v3/4/5 with VBE 2.0 disable 8x14 font which SimAnt and SimCity require to show fonts right.
Bios v1 (1.01.03, VBE 1.2) has 8x14 fonts compatibility, while newer ones do not.
Source: Re: S3 AGP Cards (and possibly others) Too Bright

Bios 2.01.16 is the latest bios released for ViRGE/DX and GX which has the brightness bug and apparently it doesn't follow the general structure of the 07 bios, according to my previous test the frequency is 45mhz 55mhz. I'm trying to tamper it to fix the brightness bug without success so far. The mclk values are located at CD-CE (still unclear) instead of C2-C3.

This is how the frequency is obtained using m, n, r: MCLK = (M+2) / (N+2) × 14.31818 / 2^R MHz
Those are the limits for the values:
135MHz < 2^R × 14.31818MHz < 270MHz
1 ≤ M ≤ 127
1 ≤ N ≤ 31
0 ≤ R ≤ 3
Apparently the fastest cards (35ns memory) are able to handle 83mhz, but it's important to check for memory errors on windows (95\98) since it's more demanding than the old dos and look for memory character corruption.

In my opinion it would be important to be able to have bios versions without the britghtness bug. The problem is that apparently noone so far was able (or willing) to fix the bios bits to correct it and the only solutions are software.

Some utilities for s3 cards:
S3VBEFIX: https://github.com/wbcbz7/S3VBEFIX
MCLK 094a to overclock s3 cards: https://www.geocities.ws/liaor2/myutil/myutil.html
MCLK Useful information: https://www.geocities.ws/liaor2/myutil/mclk.html
SVBL (Shadow Video BIOS loader), a small utility which loads video BIOS from image to shadow memory: Re: S3 AGP Cards (and possibly others) Too Bright
S3 Trio bios collection and information: https://dosdays.co.uk/topics/Manufacturers/s3_trio.php
S3 Virge bios collection and information: https://dosdays.co.uk/topics/Manufacturers/s3_virge.php

Various vogons forum topics on the matter:
S3 AGP Cards (and possibly others) Too Bright: S3 AGP Cards (and possibly others) Too Bright
How do I flash VBIOS on an S3 card? How do I flash VBIOS on an S3 card?
S3 ViRGE BIOS RAM timings S3 ViRGE BIOS RAM timings
S3VBEFIX topic: S3VBEFIX - PC\MS-DOS TSR fix for S3 VESA 2.0 video BIOS

Nemo1985 wrote on 2026-03-04, 22:14:

Bios 2.01.16 is the latest bios released for ViRGE/DX and GX which has the brightness bug and apparently it doesn't follow the g […]
Show full quote

Bios 2.01.16 is the latest bios released for ViRGE/DX and GX which has the brightness bug and apparently it doesn't follow the general structure of the 07 bios, according to my previous test the frequency is 45mhz 55mhz. I'm trying to tamper it to fix the brightness bug without success so far. The mclk values are located at CD-CE (still unclear) instead of C2-C3.

This is how the frequency is obtained using m, n, r: MCLK = (M+2) / (N+2) × 14.31818 / 2^R MHz
Those are the limits for the values:
135MHz < 2^R × 14.31818MHz < 270MHz
1 ≤ M ≤ 127
1 ≤ N ≤ 31
0 ≤ R ≤ 3
Apparently the fastest cards (35ns memory) are able to handle 83mhz, but it's important to check for memory errors on windows (95\98) since it's more demanding than the old dos and look for memory character corruption.

In my opinion it would be important to be able to have bios versions without the britghtness bug. The problem is that apparently noone so far was able (or willing) to fix the bios bits to correct it and the only solutions are software.

Hi, @Nemo1985,
I have S3 ViRGE/GX video card with BIOS version 2.01.16 with VBE 2.0 and I managed to fix the "bright bug" for it. The stock frequency for my card is 75MHz: M = 61, N = 1, K = 2 -> 75.1695 MHz -> 0x3D41. VRAM chips are SGRAM NPNX NN5283206QF-12, but I can't find any datasheet for them. It may be of use to you and others, so I'm attaching both the original and the modified (fixed) BIOS.

Modified BIOS (75MHz, no bright bug):

Original BIOS (75MHz, bright bug):

For those, who can not program and replace the BIOS chip on their S3 Trio64V2/DX (86C775), Trio64V2/GX (86C785), ViRGE/DX (86C375), ViRGE/GX (86C385), ViRGE/GX2 (86C357), Trio3D (86C365) and possibly even Savage3D, Savage4 cards, I have a software solutions for the bright bug: Software solutions for BIOS bright bug in S3 cards based on 86C775, 86C785, 86C375, 86C385, 86C357, 86C365... chips

aVd wrote on 2026-05-14, 07:13:

Hi, @Nemo1985, I have S3 ViRGE/GX video card with BIOS version 2.01.16 with VBE 2.0 and I managed to fix the "bright bug" for it […]
Show full quote

Hi, @Nemo1985,
I have S3 ViRGE/GX video card with BIOS version 2.01.16 with VBE 2.0 and I managed to fix the "bright bug" for it. The stock frequency for my card is 75MHz: M = 61, N = 1, K = 2 -> 75.1695 MHz -> 0x3D41. VRAM chips are SGRAM NPNX NN5283206QF-12, but I can't find any datasheet for them. It may be of use to you and others, so I'm attaching both the original and the modified (fixed) BIOS.

Modified BIOS (75MHz, no bright bug):

The attachment Jaton S3 ViRGE-GX-ver.2.01.16J-MOD.zip is no longer available

Original BIOS (75MHz, bright bug):

The attachment Jaton S3 ViRGE-GX-ver.2.01.16J.zip is no longer available

For those, who can not program and replace the BIOS chip on their S3 Trio64V2/DX (86C775), Trio64V2/GX (86C785), ViRGE/DX (86C375), ViRGE/GX (86C385), ViRGE/GX2 (86C357), Trio3D (86C365) and possibly even Savage3D, Savage4 cards, I have a software solutions for the bright bug: Software solutions for BIOS bright bug in S3 cards based on 86C775, 86C785, 86C375, 86C385, 86C357, 86C365... chips

Thank you very much!
After the very educative bits and bolts guides has become much easier to modify the bios to correct the bug, I modified all my bios but still didn't try them.
I checked on my excel file and my GX is clocked at 60 mhz.
I will gladly try your bios.

Nemo1985 wrote on 2026-05-14, 07:30:

Thank you very much! After the very educative bits and bolts guides has become much easier to modify the bios to correct the bug […]
Show full quote

Thank you very much!
After the very educative bits and bolts guides has become much easier to modify the bios to correct the bug, I modified all my bios but still didn't try them.
I checked on my excel file and my GX is clocked at 60 mhz.
I will gladly try your bios.

Sorry, that I saw this thread so late. Sill, it's nice to share some useful information and experience.

I can't overclock my card with this BIOS more than 79MHz without artifacts to appear. Could you share your 60MHz BIOS 2.01.16 version to compare them? Maybe the bytes for setting the VRAM timings will pop-up at last.

aVd wrote on 2026-05-14, 07:56:

Nemo1985 wrote on 2026-05-14, 07:30:

Thank you very much! After the very educative bits and bolts guides has become much easier to modify the bios to correct the bug […]
Show full quote

Thank you very much!
After the very educative bits and bolts guides has become much easier to modify the bios to correct the bug, I modified all my bios but still didn't try them.
I checked on my excel file and my GX is clocked at 60 mhz.
I will gladly try your bios.

Sorry, that I saw this thread so late. Sill, it's nice to share some useful information and experience.

I can't overclock my card with this BIOS more than 79MHz without artifacts to appear. Could you share your 60MHz BIOS 2.01.16 version to compare them? Maybe the bytes for setting the VRAM timings will pop-up at last.

I agree, sharing is very important. I just checked and I was able to overclock my card to 85 mhz (with MCLK094A) back in time.
I just checked and I noticed I didn't correct the bios of the GX, I also have two bios versions 66 and 81 mhz.

Nemo1985 wrote on 2026-05-14, 08:33:

I agree, sharing is very important. I just checked and I was able to overclock my card to 85 mhz (with MCLK094A) back in time.
I just checked and I noticed I didn't correct the bios of the GX, I also have two bios versions 66 and 81 mhz.

Thank you very much! I'll try to find the bytes related to VRAM timings setting in our ViRGE/GX 2.01.16 BIOSes.

aVd wrote on 2026-05-14, 08:41:

Nemo1985 wrote on 2026-05-14, 08:33:

I agree, sharing is very important. I just checked and I was able to overclock my card to 85 mhz (with MCLK094A) back in time.
I just checked and I noticed I didn't correct the bios of the GX, I also have two bios versions 66 and 81 mhz.

Thank you very much! I'll try to find the bytes related to VRAM timings setting in our ViRGE/GX 2.01.16 BIOSes.

Sorry I forgot to say that mclk also helps with memory settings, I remember there is a setting about 1 cycle or 2 cycle, if you have a dos rig give it a look it could prove useful.
I'm looking forward to know your reports

Nemo1985 wrote on 2026-05-14, 08:44:

Sorry I forgot to say that mclk also helps with memory settings, I remember there is a setting about 1 cycle or 2 cycle, if you have a dos rig give it a look it could prove useful.
I'm looking forward to know your reports

Yes, I do have a suitable DOS rig for experiments. Thanks for reminding me, that MCLK can read the VRAM timings set by the BIOS. I have to play with my ViRGE/GX and these BIOSes to find out which bytes could possibly contain these settings.

Looking at the three 2.01.16 BIOSes, that now I have, I just came with some 2D/3D clock frequency settings theory. I have to test it on my GX card:

According to MCLKCONV 0x3D41 gives 75MHz - the two bytes in the yellow rectangles. MCLK reports this as 75MHz frequency for my card.

In the other hand, the frequency setting bytes, that differ between the BIOSes, are the previous two bytes (in the green rectangles). From MCLKCONV: 0x4142 gives 60MHz, 0x5942 gives 81MHz and for my ViRGE/GX 0x2C41 gives 55MHz (I always had a feeling, that it is too slow in 3D even for S3 "deccelerator" it is acutally 75MHz in 3D - proven by TR tests).

The bytes that differ, but does not belong to text strings, are marked in black rectangles. Possibly one of these may be related to VRAM timings.

I have to check if "green" and "yellow" frequency bytes are related to 2D and 3D frequency settings.

P.S. No other byte differences noticed, except the last byte for the checksum, which is fine.

I have a ViRGE(325) card and these byte pairs (28 41 28 41) give 50MHz (in 3D it's even slower than my slow ViRGE/GX, 0x2841 = 50MHz according to MCLKCONV and possibly both 2D and 3D frequencies are set to 50MHz), but they are located at another address in the BIOS.

EDIT: I was wrong, that second bytes pair set 2D frequency. It turned out, that they set 3D frequency.

Found two more 2.01.16 BIOSes named for 66 and 45MHz from unknown ViRGE/GX cards. The bytes for frequencies are in strange pairs. I really have to test all of these to check my 2D/3D frequencies theory:

0x5645 ~ 45MHz
0x4842 = 66MHz (as stated in the dump file name)

0x3D43 = 45MHz (3D 43 3D 43 - probably both 2D and 3D frequencies are set to 45MHz as stated in the dump file name)

With all those pictures I'm thinking that MCLK in DOS reports only second "yellow" bytes pair - probably for 2D 3D frequency (TR1 tests revealed, that second bytes pair sets 3D frequency). And the more important for our ViRGE/GX "deccelerators" is the first second "green" bytes pair (TR1 tests revealed, that second bytes pair sets 3D frequency).

So, to me seems like yours two 2.01.16 BIOSes are set to 60MHz and 81MHz in 3D mode and mine is not 75MHz, but 55MHz in 3D mode. The last two, that I found, both must be set to ~45MHz in 3D mode (MCLK and TR1 tests revealed, that this is wrong suggestion).

As for VRAM timings setting, the main suspect is byte at address 0x3c. I have one idea how to test this too.

EDIT: I was wrong, that second bytes pair set 2D frequency. It turned out, that it sets 3D frequency.

That's a very interesting finding.
So if I got it right there are 2d and 3d frequencies they change according to what's running? It sounds very futuristic since the first cards I remember to have this feature are the Radeon 9x00 which are some years younger.
2d probably text mode in dos and then 3d anything else?
If you need me to test something just ask right away, I'm messing with my pentium 4 build but if it won't work I can put on the desk a older slot 1 build.

For now this is just my theory, because I have no other explanation for different frequency byte pairs in ViRGE/GX BIOSes. On my older pure ViRGE(325) card the bytes pairs are equal regardless if they are located at slightly different address in the BIOS.

I suspect, that MCLK in DOS only reports the frequency according to second bytes pair, and it has to be for 2D 3D mode (TR tests revealed, that second bytes pair sets 3D frequency), if my theory is valid. Good old Tomb Raider 1 (with S3 ViRGE acceleration executable - it shows FPS) and RAMBIOS 1.5 (not to reflash my card's BIOS chip 100 times) will help me check, if my suggestions are valid.

AFIK S3 ViRGE and Trio3D cards share 2D core based on S3 Trio64 (2D only "GUI accelerator" card). So, there's some theoretical possibility, that two different frequencies for 2D and 3D cores are possible.

EDIT: I was wrong, that second bytes pair sets 2D frequency. It turned out, that it sets 3D frequency.

And here are the results after testing all five 2.01.16 BIOSes with TR:

Actually the second bytes pair (in the yellow rectangle) sets 3D frequency and MCLK in DOS reports exactly this 3D core frequency.

All BIOSes with 0x3D41 second bytes pair gave me exactly the same FPS (highest) results, MCLK reports those as 75MHz, so they seem like actually are set at 75MHz in 3D mode (sorry, your "81MHz" BIOS actually works at 75MHz in 3D, but your "66MHz" is 75MHz too, and my BIOS is also 75MHz in 3D).

The BIOS with 0x4842 second bytes pair gave me slightly lower FPS result compared to the ones with 0x3D41, MCLK reports this one as 66MHz, so it seems like it is really 66MHz in 3D mode.

The BIOS with 0x3D43 second bytes pair gave me the worst FPS (lowest) result, MCLK reports this one as 45MHz, so seem like it is 45MHz in 3D mode.

So, the purpose of the second byte pair for setting the 3D core frequency question is now solved to me. I can only suggest, that the first bytes pair is for 2D frequency setting, until I find some suitable 2D (GUI) testing software.

Another day I will play with byte (VRAM timings suspect) at address 0x3c...

Thanks for the precious findings, thank you for the correction about the bios, I'm going to correct the text then.
It could be that the 2d frequencies are just used during post?
Or maybe not at all?
I'm 90% sure that in windows 98 (using powerstrip) the frequency used are the 3d ones, but I will test it later today\this evening and report back.

You're welcome! I was also curious to find out, why those bytes pairs differ and which one of them actually sets the 3D mode frequency.

As for the first bytes pair - it may be unused or used for 2D mode. I see no reason for setting different 3D core frequency during POST, but who knows... I really have to check this. Do you have any suggestion for 2D only video card tests (preferably for DOS, since I don't have any windows installed on this test rig)?

aVd wrote on 2026-05-14, 15:46:

You're welcome! I was also curious to find out, why those bytes pairs differ and which one of them actually sets the 3D mode frequency.

As for the first bytes pair - it may be unused or used for 2D mode. I see no reason for setting different 3D core frequency during POST, but who knows... I really have to check this. Do you have any suggestion for 2D only video card tests (preferably for DOS, since I don't have any windows installed on this test rig)?

I honestly struggle to understand what could be more 2d than plain text mode in dos, so we need a benchmark that measure the scrolling speed? Just characters speed? In this case I suggest Landmark System Speed Test (v6.0). I also found Vidspeed (but never tried it): https://thandor.net/benchmark/73
Another test seems to be PCBench (v9.0), but again I never tried it but this link may come handy: VOGONS Benchmark archive?

Nemo1985 wrote on 2026-05-14, 15:57:

I honestly struggle to understand what could be more 2d than plain text mode in dos, so we need a benchmark that measure the scrolling speed? Just characters speed? In this case I suggest Landmark System Speed Test (v6.0). I also found Vidspeed (but never tried it): https://thandor.net/benchmark/73
Another test seems to be PCBench (v9.0), but again I never tried it but this link may come handy: VOGONS Benchmark archive?

Thanks for suggestions! I have this Landmark test, but it only reports CPS (characters per second), so I don't know how stressful for S3 ViRGE/GX 2D core it will be. I'll check out others too.

Windows GUI is also 2D (if not using "aero" or whatever it was called in those newer versions). Or everything, that is not rendered in 3D is 2D, not only plain text mode.

And I have one request for you. Since you have test rig with windows and PowerStrip installed, could you check out the 66MHz and 45MHz 2.01.16 BIOSes too? Just to be sure, that those are really set at lower frequencies for 3D, compared to our native 75MHz S3 ViRGE/GX cards BIOSes? Get them from the attached archive:

aVd wrote on 2026-05-14, 16:07:

Thanks for suggestions! I have this Landmark test, but it only reports CPS (characters per second), so I don't know how stressfu […]
Show full quote

Nemo1985 wrote on 2026-05-14, 15:57:

I honestly struggle to understand what could be more 2d than plain text mode in dos, so we need a benchmark that measure the scrolling speed? Just characters speed? In this case I suggest Landmark System Speed Test (v6.0). I also found Vidspeed (but never tried it): https://thandor.net/benchmark/73
Another test seems to be PCBench (v9.0), but again I never tried it but this link may come handy: VOGONS Benchmark archive?

Thanks for suggestions! I have this Landmark test, but it only reports CPS (characters per second), so I don't know how stressful for S3 ViRGE/GX 2D core it will be. I'll check out others too.

Windows GUI is also 2D (if not using "aero" or whatever it was called in those newer versions). Or everything, that is not rendered in 3D is 2D, not only plain text mode.

And I have one request for you. Since you have test rig with windows and PowerStrip installed, could you check out the 66MHz and 45MHz 2.01.16 BIOSes too? Just to be sure, that those are really set at lower frequencies for 3D, compared to our native 75MHz S3 ViRGE/GX cards BIOSes? Get them from the attached archive:

The attachment 2.01.16-VBE20-BBug.zip is no longer available

Sure thing, I am going to test them and report back. I have a question, do you know a faster way to flash the bios on the card or do I need to use the eeprom writer?
If you have an utility it would come handy.
As for 2d\3d I agree with your view, but then in windows it should be in 2d all the time, while I'm pretty sure it's not. My guess is that the 2d freq are some kind of compatibility mode maybe to use with S3 Virge (S3D)?

No, I don't know a faster way to reflash your BIOS ROM chip on these S3 ViRGE video cards, if it is (E)EPROM. Usually these video cards came out of the factory with OTP-ROM chip, so no reflash is possible, until the chip is changed with compatible (E)EPROM.

But you do not need any flashing procedures to test different BIOSes for your S3 ViRGE cards. The S3 ViRGE videocard's BIOS is usually loaded in the shadow RAM during POST and after this it is used from there (OS reads faster from RAM, than from ROM chip). So, you can also use RAMBIOS 1.5 to change S3 ViRGE's BIOS on-the-fly from DOS prompt. I used RAMBIOS 1.5 for the TR tests, because my ViRGE/GX is with OTP-ROM chip, which I can not replace with compatible (E)EPROM.

Boot to DOS, load the test BIOS with RAMBIOS (you may use "MCLK /0" to check out, that the tested BIOS is loaded and active in the shadow RAM), start windows and check what PowerStrip will show.

RAMBIOS 1.5:

Shadow Video BIOS Loader 2.3 also may work, but RAMBIOS is simple and smaller in size.

And belive me, In windos everyting is 2D rendered, if you are not using windows "aero" 3D interface (I think this one is in Vista and newer) or until you start some 3D software (no matter windowed or full screen).

S3D API is not valid suggestion, as I already tested the BIOSes with S3D API in TR and when I wrote "3D mode" for S3 ViRGE, I mean exactly the native S3D API 😉 But maybe DirectX3/5 related. I didn't tested anything with Direct3D in windows.