analog_programmer wrote on 2024-04-04, 20:16:

The thread is very interesting.

Any opinions for 3DLabs Permedia 2? Can this one qualify in the category "worst fastest early 3D cards"? I got one recently, but sill have no time to test it in windows 98/xp.

Permedia 2 is really disappointing and not really even bad in interesting ways. OpenGL support is shockingly rough considering who the card is from, overall image quality is pretty spotty, and it's just flat out slow. There's no real charm or unique look like say a Matrox Mystique, it's just very forgettable.

I have a theory about the Alliance aT3D. At some point their Direct3D driver had a huge diet in size down to 1/5 of the previous dimension around the time of DX5 support. The 2D driver evolved but maybe the proprietary 3D driver did not it's possible to support mostly the 3dfx acceleration through their driver. Why supporting two Direct3D accelerators on the same card when the Voodoo Rush was the only big project having that chip. The sw and the silicon with the now "new" AT25 (maybe less good aT3D cores or 3D disabled not working units rebranded) became focused as a 2D solution in the wait of their unreleased next gen designs.

I think last drivers were not really oriented for the "old" aT3D chip but mostly for the AT25 and AT25/Voodoo Rush, which beside Glide API driver has its own Direct3D driver and at that point the aT3D became the cheap 2D solution with few 3D support the Voodoo Rush needed. But the real 3D driver for the single Directx 3/5 own accelerator it once was expected to be, IMHO was already sort of "discontinued" in the summer 1997.

I have an Alliance AT25 card that has 3D functionality, about as broken as as what I read about the AT3D.
For running Unreal on these older cards, you will usually have better luck using the original Unreal with patch 219, as opposed to Unreal Gold.

pixel_workbench wrote on 2024-04-07, 14:36:

I have an Alliance AT25 card that has 3D functionality, about as broken as as what I read about the AT3D.
For running Unreal on these older cards, you will usually have better luck using the original Unreal with patch 219, as opposed to Unreal Gold.

Thanks, I'll try the 219 patch on Unreal; I tried the final official one on Unreal Gold game and it gave the texture size error seen also launching Unreal Tournament. Imho beside what the AT25 chip was really inside, what seems broken were first the drivers that instead of getting better got "worse" probably because their market target changed temporary from 2D/3D to 2D only more or less. Of course they were low cost solutions for early 3D accelerators market. They were probably supported more as Voodoo Rush part than single chip boards.

386SX wrote on 2024-04-02, 18:09:

analog_programmer wrote on 2024-04-02, 18:04:

JustJulião, I'm not 100% sure, but as far as I know the AT25 chip is cut-down version of AT3D and it lacks 3D functionalities.

I don't have the AT25 to test that should in fact be the 2D part of the aT3D chip but I'd not be suprised if it was the same rebranded chip with just a driver modification. But of course it might have been a different core at first or maybe later during late 1997 production.

I randomly picked a AT3D/AT25 driver on driversguide (it does both). After deleting the Voodoo Rush from the device manager and installing it, it recognized an aT3D card, which was encouraging.

The result is that... the Voodoo Rush continues to be used. I confess I hadn't imagined such a scenario. 3DCC doesn't switch cards when set.

JustJulião wrote on 2024-04-07, 22:57:

I randomly picked a AT3D/AT25 driver on driversguide (it does both). After deleting the Voodoo Rush from the device manager and […]
Show full quote

386SX wrote on 2024-04-02, 18:09:

analog_programmer wrote on 2024-04-02, 18:04:

JustJulião, I'm not 100% sure, but as far as I know the AT25 chip is cut-down version of AT3D and it lacks 3D functionalities.

I don't have the AT25 to test that should in fact be the 2D part of the aT3D chip but I'd not be suprised if it was the same rebranded chip with just a driver modification. But of course it might have been a different core at first or maybe later during late 1997 production.

I randomly picked a AT3D/AT25 driver on driversguide (it does both). After deleting the Voodoo Rush from the device manager and installing it, it recognized an aT3D card, which was encouraging.
IMG20240408005444.jpg
The result is that... the Voodoo Rush continues to be used. I confess I hadn't imagined such a scenario. 3DCC doesn't switch cards when set.

Drivers seems the same for aT3D and AT25 but it's not easy to find the older ones compiled from Alliance but only the ones compiled for the Voodoo Rush which end up in different combinations with drivers from Alliance, 3dfx, OEM, clocks, driver versions changing, OEM apps and customizations.

Which are not easy to use too cause the Alliance driver into OEM packages seems customized more than thought. But having a real Voodoo Rush may be easier to work with. Do the 3dfx module is still used also for Direct3D?

386SX wrote on 2024-04-08, 03:57:

JustJulião wrote on 2024-04-07, 22:57:

I randomly picked a AT3D/AT25 driver on driversguide (it does both). After deleting the Voodoo Rush from the device manager and […]
Show full quote

386SX wrote on 2024-04-02, 18:09:

I don't have the AT25 to test that should in fact be the 2D part of the aT3D chip but I'd not be suprised if it was the same rebranded chip with just a driver modification. But of course it might have been a different core at first or maybe later during late 1997 production.

I randomly picked a AT3D/AT25 driver on driversguide (it does both). After deleting the Voodoo Rush from the device manager and installing it, it recognized an aT3D card, which was encouraging.
IMG20240408005444.jpg
The result is that... the Voodoo Rush continues to be used. I confess I hadn't imagined such a scenario. 3DCC doesn't switch cards when set.

Drivers seems the same for aT3D and AT25 but it's not easy to find the older ones compiled from Alliance but only the ones compiled for the Voodoo Rush which end up in different combinations with drivers from Alliance, 3dfx, OEM, clocks, driver versions changing, OEM apps and customizations.

Which are not easy to use too cause the Alliance driver into OEM packages seems customized more than thought. But having a real Voodoo Rush may be easier to work with. Do the 3dfx module is still used also for Direct3D?

Yes, in D3D too (Final Reality, Turok D3D).

JustJulião wrote on 2024-04-08, 07:47:

386SX wrote on 2024-04-08, 03:57:

JustJulião wrote on 2024-04-07, 22:57:

I randomly picked a AT3D/AT25 driver on driversguide (it does both). After deleting the Voodoo Rush from the device manager and installing it, it recognized an aT3D card, which was encouraging.
IMG20240408005444.jpg
The result is that... the Voodoo Rush continues to be used. I confess I hadn't imagined such a scenario. 3DCC doesn't switch cards when set.

Drivers seems the same for aT3D and AT25 but it's not easy to find the older ones compiled from Alliance but only the ones compiled for the Voodoo Rush which end up in different combinations with drivers from Alliance, 3dfx, OEM, clocks, driver versions changing, OEM apps and customizations.

Which are not easy to use too cause the Alliance driver into OEM packages seems customized more than thought. But having a real Voodoo Rush may be easier to work with. Do the 3dfx module is still used also for Direct3D?

Yes, in D3D too (Final Reality, Turok D3D).

Maybe through the Alliance chip driver it can still directly use/launch the 3dfx hw with such design, I imagine the o.s. care about the main Alliance chip and not much the 3dfx part. This weekend I did many tests but each time I can barely keep the rendering seen in the previous pages or often even having a broken vga installation; I understand the entire focus for the 3dfx ended up complicating a lot the Alliance driver itself.

After too many mixed installations with the aT3D card I had to format and reinstall again the o.s., still testing it in the i440BX board, Pentium II 400, 128MB PC100 and Win 98 First Edition updated. I tried starting from an old .203x driver package and it make Terracide demo game works again great. I'll post more screenshots and it's the best looking game running for sure, frame rate's smooth, hardware accelerated and mostly well rendered. I'd like to test the .204x drivers but OEM only 3dfx packages having those old drivers are customized and would need tweaking; I'm doing mixed test to better understand the different logic. The right spot should be from the .203x to the .2046 but drivers changed the way they worked at the time of DX5 API and Voodoo Rush early release, so it's not really easy considering they were probably not caring a lot about the aT3D/25 3D acceleration anymore after those. I'll keep this updated.

Can you confirm me that with time correct other accelerators, Viper Racing demo of 1998 is actually running accelerated and post some screenshots about it?

Viper Racing has no software renderer, it's accelerated or nothing. I expect the smoke to be unblended (assuming it gets past caps checks, I never got a PCX2 through to Viper in the day)

leileilol wrote on 2024-04-14, 01:23:

Viper Racing has no software renderer, it's accelerated or nothing. I expect the smoke to be unblended (assuming it gets past caps checks, I never got a PCX2 through to Viper in the day)

Thanks, I didn't remember if the demo or the full game had some low quality sw renderer just to be sure. The demo works well on the aT3D, very few gfx errors while a bit heavy @ 640x480 of course.

Great job 386SX, now I want to give AT3D a second look. My focus on the latest drivers is sometimes a weakness.

leileilol wrote on 2024-04-14, 01:23:

Viper Racing has no software renderer, it's accelerated or nothing. I expect the smoke to be unblended (assuming it gets past caps checks, I never got a PCX2 through to Viper in the day)

It gets past.

Putas wrote on 2024-04-14, 10:37:

Great job 386SX, now I want to give AT3D a second look. My focus on the latest drivers is sometimes a weakness.

Thank you!

It always felt strange that the chip would render that way with a final driver (and not only that version). The problem is that nowdays there are only a couple of old Directx3 drivers around and the others are a difficult territory to even install. But after two weeks of testing it became a quest to give credit to this poor low end 3D chip that was running "mostly 2D oriented drivers" partially forgetting its early driver work that probably gave problems to the 3dfx module sharing two accelerators on the same card. For the audio problems I suggest to try on the ALi chipset boards and maybe a fast time correct K6-2 /3 and maybe ISA audio cards. Both the i430 and i440 has more problems imho. I'm doing various tests anyway, later I'm going to post new screenshots. 😀

Putas wrote on 2024-04-14, 10:37:

My focus on the latest drivers is sometimes a weakness.

From my experience, the latest drivers often cause issues for these early 3D accelerators.

For example, the ATi Rage XL loses support for paletted textures when using drivers released in 2001. Same thing for the S3 Savage4. And Matrox G400 loses TurboGL with later drivers, while their ICD is still not up to pair in terms of performance.

Joseph_Joestar wrote on 2024-04-14, 11:32:

Putas wrote on 2024-04-14, 10:37:

My focus on the latest drivers is sometimes a weakness.

From my experience, the latest drivers often cause issues for these early 3D accelerators.

For example, the ATi Rage XL loses support for paletted textures when using drivers released in 2001. Same thing for the S3 Savage4. And Matrox G400 loses TurboGL with later drivers, while their ICD is still not up to pair in terms of performance.

The whole Voodoo Rush collaboration ended up retiring the early aT3D acceleration to give space to the 3dfx project. Some OEM AT25/Voodoo Rush oriented drivers really destroy the 3D acceleration not even starting in apps while still aT3D compatible; while the later (not only latest unfortunately) not-OEM aT3D/25 drivers only keep some Dx5 compatibility but mostly hardly working as well known. So it's about testing the spot where the aT3D tried itself to accelerate Directx3 and early Directx5 games without 3dfx at all. And many forgotten drivers are still lost that may render even better than I tried with various fixes, if someone ever archived the ALSC company webpages/FTP external links in that few months range I'm looking for. While forcing a bit different drivers, I even saw a good fps boost (no overclock) and some fix to overcome hw limitations.

Anyway, various newer games fails of course but not totally. G-Police Demo is actually almost rendering but there're some strange effects above 1/3 of the screen, Resident Evil actually runs but with Matrox or Rendition path can't render the main background images unfortunately.

Some more screenshots with aT3D card with previous tested apps and a bit newer driver. I'm going to test more titles but still searching for those who work a bit better.

aT3D, Final Reality again and now working Gouraud Shading, sort of.. Final reality score is 2.49 with the P-II 400. Audio sound now is really bad; it's better and much faster for the rendering too to test without it. That problem can't probably be fixed. The more CPU/VGA is busy or the more quality the audio set or the older driver it is and the whole audio will result worse or.. very worse. As said, I found a bit better situation on the ALi chipset.

Continue..

Oh another game that works well from 1998! When expectations are so low, it's impressive how fast this card was with all the rendering shortcuts it clearly take to be as fast as it was supposed to be.

aT3D and Colin McRae Rally game. Default 3D rendering path (Not the sw renderer of course, while still not tried the S3, Matrox etc.. paths), 640x480 mostly all to high details. Some few graphic errors but quite good.