stynx wrote on 2022-05-21, 15:45:

I just tested "Virtual Wings" and the Version 1.0.2 is compatible with the QuickDraw 3D accelerator. It runs pretty smoothly and […]
Show full quote

I just tested "Virtual Wings" and the Version 1.0.2 is compatible with the QuickDraw 3D accelerator. It runs pretty smoothly and has textures (that seem to be working).

Tested games with 100% support so far:
- Virtual Wings
- Havoc

Games that work with texture errors:
- Nanosaur
- Weekend Warrior

Games that don't work:
- Tomb Raider (1 &2) -> not recognized
- Quake 1 - RAVE patch -> Error 10 Freeze
- VR Soccer -> not recognized
- Mech Warrior 2 -> not recognized
- Nascar Racing -> not recognized
- Quake II -> Open GL/not supported
- IndyCar racing -> not recognized
- HotCarts -> not recognized
- Descent 1 & 2 -> not recognized
- Unreal -> not recognized

Gremlin's Virtual Pool has a QD3D version bundled with some 6500 machines "Virtual Pool 3D", it might be worth a try.
https://archive.org/details/virtual-pool

recently bought such a card...using/testing it in my 7300PPC mac OS8.

Interesting would be the comparison between the QD3D card from apple against the onboard ATI solutions apple offered back in the day? (Ati Rage Series etc.)

dr.zeissler wrote on 2023-11-18, 11:01:

Interesting would be the comparison between the QD3D card from apple against the onboard ATI solutions apple offered back in the day? (Ati Rage Series etc.)

Based on my benchmarks, i would say the Rage and Rage II would have been slower than the Apple 3D accelerator card in real-world scenarios. The main problem is that the Apple 3D card is not capable to actually compete in real world scenarios since too many features are missing. Especially the limited texture memory will make a comparison impossible but in terms of polygon-throughput it is very capable. The 3Dfx Voodoo is about 3-4 times faster than the Apple 3D card in raw benchmarks and 2-3 times faster in more realistic scenarios. A three-card setup would be able to barely match the Voodoo but the 33MHz 32bit PCI-bus is not capable managing the amount of small DMA transfers needed to fully use the potential.

A 2-card setup of QD3D accelerators would be as fast as a basic Rage Pro.

Speculation:
In terms of performance it would be capable to run Quake @640x480 in the region of 10-15fps (above Rage II levels) but since a lot of features are missing and some features like lightmap would have used up 2 rendering cycles, it could drop to Rage II levels. Since the QD3D card uses technology very similar to the Power VR, the Quake results could also be much higher since overdraw would be limited from the hardware-level. The architecture itself is quite capable and would have been a good start but the PCI-bus limitations and the artificially limited memory and texture features make it nearly unusable.

I have recently acquired a Radius Thunder 3D. The benchmark results are very disappointing. But I only have the buggy and slow 1.0.2 driver version of that card. I would like to test the card with my setup but not in this configuration.

-Jonas

Babasha wrote on 2022-05-25, 19:08:

MacOS 6 - 8.1 (are 68k only or 68k/PPC, they dont have Rave/Glide support).

MacOS 8.5 - 9.2.2 are PPC-only (they have Rave/Glide support but in native mode only, not in Classic environment under MacOS X).

I can tell for sure that 3Dfx Glide is fully accessible in Mac OS 7.5/7.6. I have both installed on my Power Macintosh 7200/75 and playing Tomb Raider in the hardware accelerated mode with a 3Dfx Voodoo Graphics.

Thank you stynx so much for all your research here!

I quite enjoyed the whole thread with all the info and your measurements / comparisons. A few notes and questions below:

S3 Virge 325 (the original one) was released in 1996, but the silicon was made in 1995 and they waited for the initial release of the Direct3D (DX2) in june (?) 1996. Many developers had the cards already in 1995 and worked on S3D-accelerated versions of their games.

It seems like the chip does some sort of hardware vertex processing from your description? Is there more than just the triangle-setup? Btw 3Dfx Voodoo, Matrox Millennium II, Rage II and (any) S3 ViRGE do not do even triangle-setup in hardware. Have you tried if the vertex processing speed is sensitive to the main CPU speed of the machine?

From my measurements and findings Rage II is slower than S3 Virge when you need to push complex geometry through the card. On the other side, S3 ViRGE is very innefficient in working with textures (especially filtered ones), so which card is faster depends on the type of a 3D scene. Btw a good ViRGE DX/GX is twice as fast as ViRGE 325 in textures - thus on par with a single QD3D card in texturing (was there any ViRGE DX/GX available for Mac? Is it possible to mod a PC version of the ViRGE to a Mac version?).

Is the texture memory of the QD3D card connected using 64bit width to the graphics chip? The texturing performance is not that bad given your mention that the texturing was added not long before the release.

Btw I see heavy inspiration from the SGI systems (if I ignore similarities to the PowerVR chips). The texturing part reminds me the SGI IMPACT architecture (including the small texture memory that is more like a texture cache, where the system expects to move textures in/out as the scene is rendered).

If there are more cards, do they slip the screen using scanline interleave (like SGI Maximum IMPACT or 3Dfx V1/V2 in SLI) or other mechanisms?

stynx wrote on 2023-12-16, 13:25:

A 2-card setup of QD3D accelerators would be as fast as a basic Rage Pro.

I have no experience with the Rage Pro on Mac, but later Win/PC “turbo” drivers leveraged full potential of the chip and it was on par with 3Dfx Voodoo Graphics - with about 50-55 Mtex/s (10x the original ViRGE 325). Also, the Rage Pro has full triangle-setup in hardware, larger texture cache and many things that made it significantly better than Rage II (just the mipmapping bug dragged the chip down in terms of visual quality).