With or without driver problems, the AT3D seems pretty useless, it seems to be missing so many features that it is even surprising that they even sold it. It looks like it could take the podium as the worst 3D card.
At least in those captures, I see the worst I have ever seen.

Hoping wrote on 2024-03-24, 16:43:

With or without driver problems, the AT3D seems pretty useless, it seems to be missing so many features that it is even surprising that they even sold it. It looks like it could take the podium as the worst 3D card.
At least in those captures, I see the worst I have ever seen.

I should test this card like as been already well done only with previous older games but the rendering problems actually get even worse easily (3DMark99 seems strangely "almost" ok). In my opinion the whole 3dfx collaboration means the 3D part of the chip would have never been optimized for 3D rendering (or maybe not able to) but mostly 2D and external 3D interface. But who knows, I still wonder if with a bigger driver package with much code to overcome so many problems how much could have done. Anyway I've tried the Hercules Voodoo Rush oriented driver package for the same chip but fail to work while it used anyway the same .2072 driver version.
2D part instead works quite well. Not so stable and bug free like many other cards, but with the Pentium MMX 233 I'm using seems working probably as good as older Alliance ProMotion 2D chips it's probably based on.

I want to find some low level old style syntethic graphic benchmarks to read how much fast it work with a basic scenario like simplest polygons and textures and how much actually "accelerate", any suggestions?

Oh, the old Half Life 1 monolights console trick works! Screenshots soon.

aT3D with Half Life Direct3D and monolights and fullbright console command; performance are almost "ok" (not really) but Fraps app seems not working here to check like 10-15fps but going on they get worse obviously in heavier scenes. Polygons are far from stable anyway, filtering is awful but at least it looks like a real 3D hardware accelerated game..

Let's speculate. It is hard to believe that the artifacts are all due to hardware bugs. Suppose that texture perspective correction is really broken. What if they tried to soften the distortion by rendering the textures at offset and dither between them? Edges are harder to recognize, masking the bad perspective to a degree. Additionally there is a chance for a sort of texture filter. If you render same texture more times at an offset, you can dither between. The 3dmark test 7 texture is a green logo on black background. Yet the AT3D renders it with something like a dither mask and many blue pixels instead of green. What if second such texture would come with yellow pixels at the right places to blend into green? When the process is not finished we get those completely bonkers colors.

I am aware how insane it would be, but maybe someone more knowledgable will explain more.

Putas wrote on 2024-03-24, 18:24:

Let's speculate. It is hard to believe that the artifacts are all due to hardware bugs. Suppose that texture perspective correction is really broken. What if they tried to soften the distortion by rendering the textures at offset and dither between them? Edges are harder to recognize, masking the bad perspective to a degree. Additionally there is a chance for a sort of texture filter. If you render same texture more times at an offset, you can dither between. The 3dmark test 7 texture is a green logo on black background. Yet the AT3D renders it with something like a dither mask and many blue pixels instead of green. What if second such texture would come with yellow pixels at the right places to blend into green? When the process is not finished we get those completely bonkers colors.

I am aware how insane it would be, but maybe someone more knowledgable will explain more.

Interesting indeed and make much sense. I wonder how disabling texture filtering would appear on this card if that would be the way this works.
It would be nice to ask someone that maybe worked in that company both driver questions, hardware limitations and how this chip ended up, I suppose should be easy to contact nowdays, maybe there's not much to know about it but it'd be great anyway. I'll upload the higher resolution specific tests images; maybe every game can be tweaked a bit for each problems. Also for the speed considerations I'm forced to try at 640x480 for a 60Mhz video chip that seems very high. If possible I'll try to force lower resolutions, 400x300 would be quite helpful for this 1997 card and to be fair while maybe thanks to this not helping final driver the card speed doesn't look the worst. The chip heat suggest it's really trying. 😁

aT3D with Final Fantasy 8 Demo (default config)

An interesting find, a Voodoo Rush driver package seems having a later 1999 Alliance driver version 2073 (!) while it may lack the 3D at all if it's the AT25 one but let's see what happen.

EDIT: it's the 2D driver build 2073, installed inside the 2072 package using only the two Alliance Promotion driver files found in that package and it get recognized by their tool as build 2073 but Windows still see it as 2072. D3D still works as usual, maybe it seems like being a bit smoother the mouse movement I should try a 2D bench. The clocks are still the same (60Mhz). I suppose I'll go back to the official driver, too bad cause it seems like compiled in the middle of 1999.

Hoping wrote on 2024-03-24, 16:43:

With or without driver problems, the AT3D seems pretty useless, it seems to be missing so many features that it is even surprising that they even sold it.

That's normal for 96-97 stuff to me. yall playing newer blendy blend games on it that'll break the whole permedia/virge/mystique/powervr1 gamut

also i'd take the strange blocky dither filtering over whatever SiS does.

If I had this AT3D i'd attempt to play stuff using techland's d3d gl driver, vertexlighting in quake3 games, mdk2 benchmark, etc.

leileilol wrote on 2024-03-25, 00:38:

also i'd take the strange blocky dither filtering over whatever SiS does.

Oh please, they are fine. Don't judge them by a few OpenGL oddities.

Putas wrote on 2024-03-24, 18:24:

Let's speculate. It is hard to believe that the artifacts are all due to hardware bugs. Suppose that texture perspective correction is really broken. What if they tried to soften the distortion by rendering the textures at offset and dither between them? Edges are harder to recognize, masking the bad perspective to a degree. Additionally there is a chance for a sort of texture filter. If you render same texture more times at an offset, you can dither between. The 3dmark test 7 texture is a green logo on black background. Yet the AT3D renders it with something like a dither mask and many blue pixels instead of green. What if second such texture would come with yellow pixels at the right places to blend into green? When the process is not finished we get those completely bonkers colors.

I am aware how insane it would be, but maybe someone more knowledgable will explain more.

Early 3D did some janky stuff to get around hardware limitations and missing functions, honestly doesnt look that bad seeing as the PS1 was in a similar boat with janky 3d and that console sold millions. Its truly amazing what you were willing to accept back in the day just to be able to play the games even if it was truly janky 3d and textures or playing at 10fps on a PC that had no business playing quake.

leileilol wrote on 2024-03-25, 00:38:

That's normal for 96-97 stuff to me. yall playing newer blendy blend games on it that'll break the whole permedia/virge/mystique […]
Show full quote

Hoping wrote on 2024-03-24, 16:43:

With or without driver problems, the AT3D seems pretty useless, it seems to be missing so many features that it is even surprising that they even sold it.

That's normal for 96-97 stuff to me. yall playing newer blendy blend games on it that'll break the whole permedia/virge/mystique/powervr1 gamut

also i'd take the strange blocky dither filtering over whatever SiS does.

If I had this AT3D i'd attempt to play stuff using techland's d3d gl driver, vertexlighting in quake3 games, mdk2 benchmark, etc.

Reading the few info it seems like giving much weight on its proprietary memory management but also its 3D engine speed. It might be marketing words but seems strange that even basic features seems broken and maybe needed to be programmed in a different way to look better. Of course it is indeed a late built early accelerator but it may have had some interesting features.

OpenGL wrapper seems difficult to run here at least trying Half Life it seems there're problems starting maybe with the "fixed" 640x480 intended resolution. Also the time correct Pentium MMX seems a bit slow maybe I should go back to the i440 and the slowest P-III.

Trashbytes wrote on 2024-03-25, 03:28:

Putas wrote on 2024-03-24, 18:24:

Let's speculate. It is hard to believe that the artifacts are all due to hardware bugs. Suppose that texture perspective correction is really broken. What if they tried to soften the distortion by rendering the textures at offset and dither between them? Edges are harder to recognize, masking the bad perspective to a degree. Additionally there is a chance for a sort of texture filter. If you render same texture more times at an offset, you can dither between. The 3dmark test 7 texture is a green logo on black background. Yet the AT3D renders it with something like a dither mask and many blue pixels instead of green. What if second such texture would come with yellow pixels at the right places to blend into green? When the process is not finished we get those completely bonkers colors.

I am aware how insane it would be, but maybe someone more knowledgable will explain more.

Early 3D did some janky stuff to get around hardware limitations and missing functions, honestly doesnt look that bad seeing as the PS1 was in a similar boat with janky 3d and that console sold millions. Its truly amazing what you were willing to accept back in the day just to be able to play the games even if it was truly janky 3d and textures or playing at 10fps on a PC that had no business playing quake.

In my opinion the problem seems like the driver didn't really try that much to get around all time correct requirements and probably missing the future support most company wanted. I don't know how much space this architecture would have to fix something but the driver package seems very light for a 96/97 card. Fixing problems in a single game would have maybe solved problems in others too. Last driver being early 1998 seems a bit soon for a card mostly seen in the 1997.

I will test anyway some other games and wrappers.

leileilol wrote on 2024-03-25, 00:38:

That's normal for 96-97 stuff to me. yall playing newer blendy blend games on it that'll break the whole permedia/virge/mystique/powervr1 gamut

also i'd take the strange blocky dither filtering over whatever SiS does.

If I had this AT3D i'd attempt to play stuff using techland's d3d gl driver, vertexlighting in quake3 games, mdk2 benchmark, etc.

Back then the first time I completed FF8, (my favorite Final Fantasy by the way), I completed it on an MMX200 with the Mystique, the Mystique is not compatible with FF8, so I played with the software renderer, and it was quite slow by the way, but it seemed playable to me at the time, and it looked better than the AT3D screenshots, if the menu looks like that it's impossible to play it.
This case seems quite exaggerated to me.
Some games that I'm curious about if they would work on the AT3D and what they would look like are the first three Resident Evil, on the Mystique they worked quite well, I'm not totally sure about the third game, but I'm pretty sure that I played the first two with the Mystique.

Hoping wrote on 2024-03-25, 11:44:

Back then the first time I completed FF8, (my favorite Final Fantasy by the way), I completed it on an MMX200 with the Mystique, […]
Show full quote

leileilol wrote on 2024-03-25, 00:38:

That's normal for 96-97 stuff to me. yall playing newer blendy blend games on it that'll break the whole permedia/virge/mystique/powervr1 gamut

also i'd take the strange blocky dither filtering over whatever SiS does.

If I had this AT3D i'd attempt to play stuff using techland's d3d gl driver, vertexlighting in quake3 games, mdk2 benchmark, etc.

Back then the first time I completed FF8, (my favorite Final Fantasy by the way), I completed it on an MMX200 with the Mystique, the Mystique is not compatible with FF8, so I played with the software renderer, and it was quite slow by the way, but it seemed playable to me at the time, and it looked better than the AT3D screenshots, if the menu looks like that it's impossible to play it.
This case seems quite exaggerated to me.
Some games that I'm curious about if they would work on the AT3D and what they would look like are the first three Resident Evil, on the Mystique they worked quite well, I'm not totally sure about the third game, but I'm pretty sure that I played the first two with the Mystique.

The menu is probably cause the 8-bit textures feature not working but detected. It's anyway very slow and resolution too high anyway beside graphic errors.

PC Player Direct3D benchmark finally is one test that seems fully rendering.. 😁.. 640x480 16bit > 10fps, 11.6fps without transparent textures, 14fps without Z-Buffer. While the software renderer using the same features looks so much better (2fps) in rendering quality. It's so interesting how strangely this chip render everything. But at least the card CAN render at lower resolution, this benchmark can force all the resolutions you want and still the board has to render it. 320x240 @ 17fps with peaks at 27fps. Interesting fact, disabling the bilinear filtering doesn't change the frame rate! Seems disabling the features in the bench actually change the graphic but doesn't increase the frame rate beside transparent textures and Z-Buffer. I'm also trying an old Direct3D bench called Direct3D Benchmark and there're test options to enable and disable features. Disabling Bilinear Filtering and going to Point Filter seems to correct the broken graphic of the sphere rendered. I'll try to make some screenshots. WizMark Lite run at 10fps and report the message it doesn't support sub-pixel correction. But seems to show the texture filtering feels like some faster idea to solve two texture problems at the same time.

I'm also trying Half Life tweaks to better understand how to stop the crazy filtering and polygons unstable. I've disabled the high quality GL_TEXTUREMODE but still seems to be far from stable.

Those reference programs sound interesting, I have also ever thought if there was something besides 3DMARK for Direct3D, for OpenGL, there is GLQuake, but for Direct3D, I didn't know of any.
Because 3DMARK99 already seems very modern to me for pre-1998 graphics cards like the ones we discussed in this thread. And also very modern for a Socket 7.
Thinking about games that used early versions of Direct3D, Hellbender and VirtuaCop come to mind, if I'm not mistaken they used DirectX 3. These can give good results with AT3D.

Back in the olden days before 3DMark, we'd drive the mule up into town, a teaming metropolis of 380 souls, and get a 5 gallon can of 'lectric currant sauce at the general store, and haul it back, then spend the night burning it up running Final Reality on th'old Pentium to gets our benchies.

Hoping wrote on 2024-03-25, 21:26:

Those reference programs sound interesting, I have also ever thought if there was something besides 3DMARK for Direct3D, for OpenGL, there is GLQuake, but for Direct3D, I didn't know of any.
Because 3DMARK99 already seems very modern to me for pre-1998 graphics cards like the ones we discussed in this thread. And also very modern for a Socket 7.
Thinking about games that used early versions of Direct3D, Hellbender and VirtuaCop come to mind, if I'm not mistaken they used DirectX 3. These can give good results with AT3D.

In fact I was thinking to use these to see where and how the aT3D miss something. I imagine the texture filtering is not working as it should probably to keep rendering as fast as possible and solve other usual early texturing problems. At this point I'd disable filtering in all games to at least increase sharpness. Half Life with nearest texture mode is at least a bit sharper.

My vote goes to the SiS 6326. It was slow, but image quality and feature support was pretty good. Can run Unreal at 320x240. A couple of site dedicated to vintage hardware have images of this board allegedly running Quake III using a MiniGL of sorts. So it's probably not the worst fastest but the best slowest 😁