You did not specify which driver settings you used. "Alpha Blending" and "3D Filter Quality" are the ones you need to look at with V3-5. Changing around those might give varying results. For V1/V2 I believe the variable that controls the filter is "SST_VIDEO_FILTER_DISABLE" and possibly others.

Also I see that 4xFSAA comes out not nearly as nicely here as it is with actual gameplay. The way it smoothens the image works well against mipmap banding/texture shimmering, so it makes up for the lack of anisotropic filtering on these cards to some extent. It provides a better level of quality than an equivalent level of MSAA would.

Nice! I'd love to know what settings you used 😀.

We should put this up on the wiki !

Ok I added a line for driver and settings.

I'd like to know if those images are reproducable, i.e. did you capture a few screenshots for each card and did they all have the same characteristics? Some of the differences seem quite random, for example V4 vs. V5 singlechip. I don't see why Voodoo4 should be blurrier when they both have the same VSA-100 chip. V5 singlechip vs. V5 SLI is weird too. If there is someone here with deeper knowledge of the hardware I'd like to know the technical reason for that, if there is any.

V2 comes out as the clear loser in this comparison. It's blurry, loses a lot of detail in those dark spots and has those arbitrary dithering spots going on. Then again some of this might be fixed via driver/variables, and perhaps most of the differences in color temperature/brightness are driver related.

d1stortion wrote:

I'd like to know if those images are reproducable, i.e. did you capture a few screenshots for each card and did they all have the same characteristics? Some of the differences seem quite random, for example V4 vs. V5 singlechip. I don't see why Voodoo4 should be blurrier when they both have the same VSA-100 chip. V5 singlechip vs. V5 SLI is weird too. If there is someone here with deeper knowledge of the hardware I'd like to know the technical reason for that, if there is any.

V2 comes out as the clear loser in this comparison. It's blurry, loses a lot of detail in those dark spots and has those arbitrary dithering spots going on. Then again some of this might be fixed via driver/variables, and perhaps most of the differences in color temperature/brightness are driver related.

The rendering characteristics do not change if I walk to some other location / map in the game. I checked other locations but decided in the end that the vortex rikers location already shows the characteristics well.
For me Voodoo4 vs Voodoo5 Singlechip does not look blurrier. It looks the same except a slight difference in color saturation, which might be due to the analog output stage of the cards.

Well compare the B2 sign on V4 16bit to V5 singlechip, the texture comes out slightly more crisp on Voodoo5.

What does this do?
SST_VIDEO_24BPP "Enable 24-bit video mode"

It's available even for Voodoo1.

d1stortion wrote:

Well compare the B2 sign on V4 16bit to V5 singlechip, the texture comes out slightly more crisp on Voodoo5.

Hmm:

Well, maybe I am growing old hehe, but I can't say that V5 is sharper than V4.

Ok for 16 bit there is a difference:

Maybe I am getting very old then because it all looks the same and great to me 😀

Interesting! Are you able to test using different cards of the same type? I've noticed image quality differences just swapping out cards in my rigs before - though never performed a detailed test like this. I ask mainly because V2s always seemed discernibly better to me than V1s but the results here pretty much refute that..

Added Grabs for
SST_VIDEO_24BPP=1
SST_VIDEO_FILTER_DISABLE=1
SSTV2_VIDEO_24BPP=1
SSTV2_VIDEO_FILTER_DISABLE=1
and ScreenSmoothing=False in Unreal

only disabling the dither filter had a real effect, image gets sharper, but you can enter the grid now...

From the Voodoo Graphics tech ref,

Bit(8) of fbzMode enables 16-bit color dithering. When enabled, native 24-bit source pixels are dithered into 16bit RGB color values with no performance penalty. When dithering is disabled, native 24-bit source pixels are converted into 16-bit RGB color values by bit truncation. When dithering is enabled, bit(11) of fbzMode defines the dithering algorithm -- when bit(11)=0 a 4x4 ordered dither algorithm is used, and when bit(11)=1 a 2x2 ordered dither algorithm is used to convert 24-bit RGB pixels into 16-bit frame buffer colors

bit truncation is crispy I guess.

24 bit definitely does something but it is probably dependent on the scenario. I also wonder if dither needs to be disabled.
http://forum.beyond3d.com/showpost.php?s=a94f … 66&postcount=40

elianda wrote:

Voodoo1 with 8 MB, so I could set 800x600 resolution.

Is there such a thing as an 8mb Voodoo 1? We need pics.

sliderider wrote:

elianda wrote:

Voodoo1 with 8 MB, so I could set 800x600 resolution.

Is there such a thing as an 8mb Voodoo 1? We need pics.

Well, seems no one checks my hardware gallery 😉.
So here are direct links:
http://mail.lipsia.de/~enigma/neu/pics/3dfx_v … magic3dplus.jpg
http://mail.lipsia.de/~enigma/neu/pics/3dfx_v … 3dplus_back.jpg

I currently try to check how 24bpp and the dithering filter play together. I could reproduce the slight color banding effect in the UT menu, but I have difficulties to see a difference in game.

edit: Found the problem. It is enough that the Key exists in the registry, independent of the actual DWord Value.

I don't know if you're interested, but here's the save game saved on the spot where I did my 640x480x16 Voodoo dithering comparison shots in Quake2 version 3.20. Extracts to baseq2/save/

Well, I also tested with a Voodoo2 combination of Dithering Filter and 24bpp setting:

First of all a check with a color gradient to see if there is any effect o 24bpp:
24BPP not set

and 24BPP set

The slight color band visible before are gone.

Now four stills of the combinations:
24BPP off, dither filter off

24BPP off, dither filter on

24BPP on, dither filter off

24BPP on, dither filter on

Well, the difference for the dithering filter is visible, but whether 24bpp is on off is difficult to distinguish.

To generate a difference image I used photoshop, put the other image as layer on top and did the difference of both.
First the difference of the same setting, to get an impression of the analog noise of the VGA signal (24BPP on, Dither Filter On, two grabs):

The main effect is that brighter pixels show a bit larger noise, which is not unexpected.

The difference of the images with dither filter off:

and with dither filter on:

Well either 24BPP is not in effect here or the effect is smaller than the average VGA noise.

In the ftp folder are also shots from UT with a difference image.

Quake2 for Leileilol

Thanks for investigating all of this. That 24bit setting is mysterious...

According to http://svn.netlabs.org/repos/odin32/trunk/src … vg/init/video.c
the SSTV2_VIDEO_24BPP variable sets bit 22 of the fbiInit1 register.
Enable 24 Bits-per-pixel video output (1=enable). Default is 0.

But there is also bit 2 of fbiinit5 which is read only:
dac_data output width (0=16-bit, 1=24-bit). Default value is the value of fb_addr_b[2] at the deassertion of pci_rst. Read only.
This seems to be basically a readback of the DAC type.
There is also bit 25 that enables 24 bit output to the DAC.

Looking at the DACs:
ICS5342 mostly on Voodoo1, can do 24 bit per pixel modes up to 640x480 at 70? Hz.
TVP3409-170CFN mostly on Voodoo2, can do 24 bit per pixel unpacked up to 800x600 at 75 Hz, and 24 bit packed mode at 1024x768 67 Hz.

So the framebuffer of the Voodoo2 itself is 16 bit only. The only hint where 24 bit plays a role is the gamma correction:
"Voodoo2 Graphics uses a programmable color lookup table to allow for
programmable gamma correction. The 16-bit dithered color data from the frame buffer is used an an index into the
gamma-correction color table -- the 24-bit output of the gamma-correction color table is then fed to the monitor or
Television."

In detail the gamma lookup on the Voodoo2 works like this:
"The Chuck internal Color Lookup table is used for gamma correction of 16-bit RGB values during video refresh.
The 16-bit RGB values read from the frame buffer are used to index into the internal video Color Lookup table.
The output of the video Color Lookup table is then fed to an external DAC. The video Color Lookup Table is stored
internally as a 33x24 RAM. As RGB values are input from memory, the 5 MSBs of a particular color channel are
used to index into the Color Lookup Table. The 3 LSBs of a particular color channel are then used to linearly
interpolate between multiple video Color Lookup Table entries. As a result of the linear interpolation performed,
smooth transitions from one Color Lookup Table index to surrounding indices results. Using linear interpolation, a
much smaller video Color Lookup Table (33 entries) can be used instead of a full Color Lookup Table (256 entries).
As a result of the linear interpolation, however, all entries stored in the videoColor Lookup Table must be
monotonically increasing."

I hope this clarifies the 24 bit setting a bit.