swaaye wrote:

One has to wonder if S3 had put effort into porting games to S3D if Virge would have been more successful in the eyes of gamers. Verite V1000 is almost as useless for D3D and OGL as Virge-arch chips after all, but Rendition put a lot of effort into getting popular games supported.

On top of that, there's also the issue of buggy DirectX and nonexistent OpenGL drivers, especially early on. (it seems there were some semi-decent -or at least more compatible- D3D drivers late in the Virge/DX/GX's life)

From what I can gather, ATi's RAGE drivers of the time weren't much better than S3's, but they certainly got many more RAGE-specific ports than S3 did. Plus, they got a relatively fast card out (with the Rage Pro) where the ViRGE line stagnated in the performance range of the Rage II.

On the note of S3D-specific ports, it also would have been good to promote use of the ViRGE's strengths, like use of 32-bit color depth (which was only moderately slower and much better looking -smooth, undithered shading/blending) as well as stress for inclusion of decent detail setting options for decent user flexibility. (since the ViRGE runs fairly decently with full features and truecolor at low screen resolutions -like 320x240 or 400x300, or also runs pretty well at higher resolutions with texture filtering disabled)

With full features enabled at 32-bit color, the ViRGE potentially has better visual quality than the Voodoo (or Verite -let alone Rage), though to keep-up speed-wise you'd have to drop the resolution, so an obvious trade-off. (or, for games catering fairly well to -if not specifically intended for- unfiltered textures, opting for lower visual quality at high resolution might be the more competitive option against the Voodoo as well -and still with considerably better visual quality than the Mystique due to proper translucency/alpha blending support)
Oh, plus there's the (non-rendering-related) issue of the Voodoo I's tendency for mediocre analog video output with issues with sharpness and/or blotchy bar/line artifacts in the output. (evident in the your recordings too)

Optimization for S3's special 4-bit (non-paletted) texture formats might also have been significant. Albeit, support for more typical 4-bit paletted textures would have been more useful. (be it either using an offset to select 15/16 colors from the on-chip CLUT, or pointing to external 15/16-color tables and quickly loading per-texture)
It's rather odd they went with that proprietary interpolated 4-bit color scale format (2 indexed color values with 15 shades/hues interpolated between) rather than just implementing a paletted 4-bit texture system. (or perhaps adding the special interpolated 4-bit format in addition to 16 color palette support, but on its own, the interpolated shades are only useful in specialized situations, like certain types of terrain, water, or sky textures)

kool kitty89 wrote:

From what I can gather, ATi's RAGE drivers of the time weren't much better than S3's, but they certainly got many more RAGE-specific ports than S3 did.

Re: 3D accelerated games list (non 3dfx/D3D/OGL)
No, they were pretty much equal, S3 actually had a tad more. Though some ATI games almost requires the PRO card, like Formula 1, Turok, and Tomb Raider.

It would take very detail image quality analysis to determine which chip of the three mentioned produce better pictures. I don't think anyone is willing to do it. And even then would remain factor of individual sensitivity to particular deficiencies.

Putas wrote:

It would take very detail image quality analysis to determine which chip of the three mentioned produce better pictures. I don't think anyone is willing to do it. And even then would remain factor of individual sensitivity to particular deficiencies.

When properly configured for 32-bit rendering (as seems to be the case with Incoming), I'd think the ViRGE would have a significant edge over the Voodoo in 16-bit color with the smoother shading and blending possible in the higher color depth. (talking purely visual quality, not speed -though the latter would be much less of an advantage for the Verite)

At 16-bit color, the dither and alpha blending artifacts of the ViRGE would put it at a disadvantage though. Given the rather mediocre improvement in smoothness achieved through dithering in these cards (relative to the plain undithered Voodoo or Verite), it seems like that feature on the Rage and ViRGE (or Riva for that matter) was a bit moot. Perhaps it's better suited to smooth shading of untextured surfaces and fails more on heavily textured areas (the relatively primitive software render dithering used in X-Wing and Tie Fighter works rather well). The dithering implementation in the Playstation also seems to fare better, especially considering the low resolutions commonly used. (dithering is also software selectable, so developers could choose whether a game benefited or not)

The edge in visual quality of the ViRGE over the Rage could be more direct, as the ViRGE tends to fare better at 16-bit color, at least in cases making the Rage's lack of filtered translucency effects obvious (ie relatively low res translucent textures in use making the problem obvious). For cases with fairly high res translucent textures, plain translucent polygons (untextured), or cases with little to no use of translucency at all, the Rage's disadvantage wouldn't be apparent either.
Plus there's still the issue of comparing 32-bit color performance, or lack thereof, depending which cards are compared.

vetz wrote:

kool kitty89 wrote:

From what I can gather, ATi's RAGE drivers of the time weren't much better than S3's, but they certainly got many more RAGE-specific ports than S3 did.

Re: 3D accelerated games list (non 3dfx/D3D/OGL)
No, they were pretty much equal, S3 actually had a tad more. Though some ATI games almost requires the PRO card, like Formula 1, Turok, and Tomb Raider.

Huh, I'd thought the Rage had a fair bit more support there for some reason. Interesting to note that some were Pro-specific games whereas several S3D games had the opposite problem of not working properly with the ViRGE DX/GX/MX cards.

On the note of GPU-specific ports, even with the overall limited number of games it's a bit odd that the Rage and ViRGE (or perhaps Mystique) didn't include a couple of the hottest titles of the time, especially Quake, or at least a specialized MiniGL driver. (albeit, in the case of the Rage -and Riva- full OpenGL drivers became available later on -another thing the ViRGE continued to lack . . . though several newer high-performance cards -like the G200 and G400- had poor/nonexistent OpenGL support for quite a while too)

Color precision cannot be reduced to comparison of basic functional parameters and it seems to me you did not yet got even those right. Both Voodoo and Vérité are dithering into 16 bit. From what I've seen I would put my money on V1000. Rage is certainly not an IQ contender.

kool kitty89 wrote:

(albeit, in the case of the Rage -and Riva- full OpenGL drivers became available later on -another thing the ViRGE continued to lack . . . though several newer high-performance cards -like the G200 and G400- had poor/nonexistent OpenGL support for quite a while too)

I had a Elsa Victory Erazor and I think they had a rather good OpenGl driver already very early. (ELSA AG was a popular german company for graphics solutions and modems)
I'am not sure if there was any interaction between Elsa and NVidia in the field of software development at this time.

Putas wrote:

Color precision cannot be reduced to comparison of basic functional parameters and it seems to me you did not yet got even those right. Both Voodoo and Vérité are dithering into 16 bit. From what I've seen I would put my money on V1000. Rage is certainly not an IQ contender.

Hmm, looking at the 3DFX screenshots on your site, I can see the dithering now . . . it's certainly much more subtle than with the Rage/ViRGE/Riva and blends much better. (which is why I assumed there wasn't dithering at all -you need good lossless screenshots -or really high quality jpg captures- to see it clearly)

Actually, that dithering looks different (and less noticeable) than many/most later implementations (ie on relatively modern video cards). It's actually really nice looking, and it's no surprise I missed it before.
It doesn't look like ordered (Bayer) dithering or error diffused dithering (floyd-steinberg, etc), but something else. Actually, it looks like it may be intended to blend smoothly after a single horizontal and vertical pass of filtering. (which particularly reminds me of the dithering used by several early Digital Pictures Sega CD FMV games -like Night Trap, Sewer Shark, Prize Fighter, and Ground Zero Texas)
And indeed, the dither pattern totally disappears from those screenshots after a single pass of h/v blur or scaling to 1/2 v/h resolution. (which is also the case for the dither used in those old sega CD games) Aside from applying actual filters, this means that the eye will tend to blend the dithering more readily as well. (especially with higher resolutions and/or smaller screens)

With the relatively subtle and restricted dithering used, it probably puts a bigger limit on smoothing of shading/blending (ie lower threshold, more posterized, or closer to undithered), but that definitely seems to be a good compromise given the results.
In some respects it reminds me of the relatively simple (and fine) dither patterns used in DOS X-Wing and Tie Fighter. (albeit those were even more restricted -namely simple checkerboard patterns)

3dfx also uses RAMDAC post processing to improve image quality, which screenshots won't capture because these frames haven't gone through the RAMDAC yet. I'm not sure if Voodoo 1 does it, but Voodoo2 does and Voodoo3 improved upon it. Voodoo2 actually displays artifacts in some cases (I've never noticed, but it is mentioned in articles).

AFAIK, Voodoo3 can do a 2x2 box or 4x1 line filter and this is configurable in the control panel. Voodoo2 does the line filter.

This is what they called 22bit color depth.

BTW, revealing 3dfx docs!
http://darwin-3dfx.sourceforge.net/voodoo2.pdf
ftp://ftp.ulakbim.gov.tr/NetBSD/misc/macallan … oodoo3_spec.pdf (includes a avenger vs banshee section)

Hypersnap can capture post filter image from Voodoo2 up.

Aha, so it does the RAMDAC filter in software on the grabbed framebuffer?
Still the easiest way is to capture directly from VGA out.

swaaye wrote:

3dfx also uses RAMDAC post processing to improve image quality, which screenshots won't capture because these frames haven't gon […]
Show full quote

3dfx also uses RAMDAC post processing to improve image quality, which screenshots won't capture because these frames haven't gone through the RAMDAC yet. I'm not sure if Voodoo 1 does it, but Voodoo2 does and Voodoo3 improved upon it. Voodoo2 actually displays artifacts in some cases (I've never noticed, but it is mentioned in articles).

AFAIK, Voodoo3 can do a 2x2 box or 4x1 line filter and this is configurable in the control panel. Voodoo2 does the line filter.

This is what they called 22bit color depth.

BTW, revealing 3dfx docs!
http://darwin-3dfx.sourceforge.net/voodoo2.pdf
ftp://ftp.ulakbim.gov.tr/NetBSD/misc/macallan … oodoo3_spec.pdf (includes a avenger vs banshee section)

I was aware of the filtering and approximated "22-bit" colordepth on Voodoo3, but I wasn't aware of the specific dithering techniques used, or that the Voodoo1 used dithering at all.
Captures from VGA out would obscure the raw dithering (or pixel-level image in general) and more so with filtering enabled, as would any lower quality lossy screenshots.
Actually, the same is true for the aforementioned dithered Sega CD games, as the composite video artifacting acts as a horizontal blur filter, though it doesn't affect the vertical. (so you end up with horizontal line/span artifacts where the dither isn't fully blended)

I haven't seen any other cards using dithering like the Voodoo (at least as depicted by Putas's pics). I've seen plenty using common ordered dithering and error diffused methods as well, but not like what I'm seeing here. Unless the Verite uses a similar method . . . but that seems to use a different method, though similarly subtle. It doesn't seem to blend completely after a single pass of h/v blur, but it does blend completely (no trace of dither) with 2 horizontal and 1 vertical blur pass (or scaling 1/4 H and 1/2 V). Albeit, specific cases of ordered/bayer dithering can exhibit similar properties.
It's definitely much finer/more subtle than the obvious ordered dither used in the Rage/Riva/ViRGE/etc in 16-bit color)

Some games in Putas's V1000 gallery seem to have no dithering at all (like Battlezone). Perhaps some games disabled dithering by default.

Dithering influenced by application is usually Gouraud shading smoothing. It may affect final image but is not really related to frame buffer dithering.

elianda wrote:

Aha, so it does the RAMDAC filter in software on the grabbed framebuffer?

I am not sure how it works.

I was thinking on the Verite's performance comparison again, and realized I'd missed checking one area:

What resolution is 3DMark being run at for the V1000? For that matter, what is the Rage or ViRGE being run at?
Since the Voodoo1 would be limited to running at 640x480 (instead of the default 800x600), the comparison would be rather skewed if the other cards weren't run at similar settings. (especially other cards of the same era, where resolution could make a huge difference)

I ran 3DMark on the V1000 and Virge GX at 640x480. My capture card usually doesn't work properly at lower resolutions.

BTW in my videos I almost always tried to capture the settings screen of whatever program I was running.

So I captured some stereoscopic 3D footage from the ELSA Revelator driver running on Riva TNT: http://www.youtube.com/watch?v=pv-KV-e9M74

It shows XL-R8R in 640x480x16.
Could someone that has also a 3D-TV or monitor give some feedback if the Youtube video shows correctly?
(You have to choose the appropriate 3D view using the "3D" button near the resolution button)

Here's another attempt at Final Reality PowerVR PCX2 footage.
http://www.youtube.com/watch?v=TrUiPfFBd_g

Here is a little bit 4x FSAA footage from V5 5500:
Dethkarz: http://www.youtube.com/watch?v=4FefVUfZbso
NFS5 Porsche: http://www.youtube.com/watch?v=v7qPK7raGiw

I didn't make this one public because my computer has been recovered and I didn't quite configure my sound properly

I've made some unlisted video captures for different posts around here on Vogons. Thought I could gather them all up.

NASCAR Racing on the NV1 - Pentium 166mhz vs Pentium 3 1050mhz - http://www.youtube.com/watch?v=Q2_6O01oC2E (Yes, the P166 is noticably quicker!)
Hi-Octane on Creative 3D Blaster PCI Pentium 133mhz - http://www.youtube.com/watch?v=ePQqqiS69Mw
Virtual On: Cybertroopers on PowerVR PCX2 - http://www.youtube.com/watch?v=39Ny-6lmi_4
Mechwarrior 2 on ATI Rage 3D Pro - http://www.youtube.com/watch?v=O2z_pyeX0uU (I made this video before my Mechwarrior 2 comparison video)
Mechwarrior 2 on the S3 Virge GX2 - http://www.youtube.com/watch?v=Mag6bBV6mIM (I made this video before my Mechwarrior 2 comparison video)
Screamer on S3 Virge GX2 - http://www.youtube.com/watch?v=x7yqJYB3dxs

For my other public videos, please see my Youtube channel: http://www.youtube.com/user/RetroCompaqGuy/videos?view=0

I never knew Hi-Octane and Nascar Racing 1 could be 3D accelerated. 😮