Hi there mrau.

Your old system was actually faster than this! Do keep in mind that this is a Pentium Classic 200, not an MMX, so your K6-2 350 should have been faster, especially if you were using the 100MHz FSB. He may have slightly more RAM than you did, although 64MB vs 80MB should for all intents and purposes be nearly identical, however if you were using 100MHz RAM it would have been faster than 66MHz EDO. On top of all this, the Riva 128 should actually be faster than the Voodoo 1, not by a whole lot mind you. Plus, we don't know what settings he used.

I'm pretty sure Quake 3 has some special detection for 3Dfx Voodoo 1 and 2 cards, but I can't seem to remember what that does. If anybody remembers, perhaps they could pitch in.

When you refer to opensource versions, do you mean sourceports? If so, I don't think they will be faster, in fact I'm pretty sure they will be quite a bit slower 😁.

indeed, i did not think of fsb and ram clock :> although amd cpu were pretty bad at this 😀
to my knowledge voodoo did some small part of the geometry setup that most other cards did not, therefore a speedup was almost guaranteed; ill admit i dont remember what part of geometry riva could do but i would really like to see if a geometry accelerating unit as in geforce or that first card that ever came out could make this fluent;

back in the day i used an addon called maxfps sometimes that gave some 30-50% in quake, of course things looked ugly as the guy in the mirror 😁

AFAIK, the source released by ID uses different (slower) libraries for sound and ports may be slower. Also, modern ports may use modern libraries (DirectX, OpenGL) that may not be available on your CPU/OS/VGA or perform slower.

I did some testing few years ago and it seems the Pentium P54 and Voodoo 1 are both too slow for Q3A.

Q3A medium details 640x480:

P200 (3x66) + V1 = 12.3 fps
P200 (2x100) + V1 = 14.6 fps
P200 (2x100) + V2 = 17.5 fps
PPro 200 + V1 = 18.4 fps
PPro 200 + V2 = 24.9 fps
K6-2 300 (2x100) + V1 = 19.6 fps
K6-2 300 (2x100) + V2 = 29.7 fps
PII 300 (3x100) + V1 = 20.2 fps
PII 300 (3x100) + V2 = 37.9 fps
PII 450 + V2 = 42 fps
PII 450 + V3 3000 = 47.7 fps

That looks like r_picmip 0 which will absolutely choke a Voodoo1 (and a Voodoo2 in some spots) so yes.

Also the source ports out there aren't faster, hell they regress entire platforms for darned reasons.... Maybe except for mine 😜

F2bnp wrote:

I'm pretty sure Quake 3 has some special detection for 3Dfx Voodoo 1 and 2 cards, but I can't seem to remember what that does. If anybody remembers, perhaps they could pitch in..

Its detection is to limit the selectable resolutions, use their gamma GL extension and prevent trilinear filtering on Voodoo2 due to some multitexturing driver bugs

havli wrote:

I did some testing few years ago and it seems the Pentium P54 and Voodoo 1 are both too slow for Q3A.

Yes, the Voodoo 1 is just way too slow. I just remembered this great video from swaaye, using a Tualatin at 1050MHz:
https://www.youtube.com/watch?v=mXpoRJjsr-g

leileilol wrote:

Its detection is to limit the selectable resolutions, use their gamma GL extension and prevent trilinear filtering on Voodoo2 due to some multitexturing driver bugs

Nothing of real interest then. 😁 Do you have any photo examples of trilinear filtering on the Voodoo2? I seem to remember other games warning against using it (Blood 2 + Shogo) as well.

Zup wrote:

AFAIK, the source released by ID uses different (slower) libraries for sound and ports may be slower. Also, modern ports may use modern libraries (DirectX, OpenGL) that may not be available on your CPU/OS/VGA or perform slower.

I remember ioquake 3 having terrible performance on the library computers at my old high school. 😜 I believe most of those boxes were Pentium 4 2.8GHz Northwoods with Geforce FX 5200s. They ran the normal Q3 client just fine, as well as Medal of Honor: Allied Assault, another Q3 engine game.

F2bnp wrote:

Do you have any photo examples of trilinear filtering on the Voodoo2? I

Yeah, though this is shown through r_colorMipLevels and a MesaFX ICD (to get around q3's v2 trilinear check, as well as vertex lighting so trilinear can actually work)

It can also dither mipmaps together. This following screenshot is taken on a V3, same technique used (I couldn't get it working on my v2 driver for some reason)

Damn, thanks for illustrating this. Is this an issue with every game that uses multitexturing on the V2? Can it do trilinear on older, non multitextured titles properly?

Also, what's causing this, hardware or driver issue and if the latter, why didn't 3Dfx ever fix it?

This is the latest of the latest official 3dfx drivers......so nope. Voodoo3's 3dfx tools help file mentions the multitexture limitation on trilinear as well

EDIT: Okay here's some VOODOO2 quake3 1.11 demo screenshots with MesaFX playing with some filtering settings. something available to the public to reproduce at least 😜

Dithered mipmaps are enabled through the V2 control panel in the Advanced>Glide section and doesn't seemt o matter whether you use trilinear or bilinear settings in a game

Turning it off allows true trilinear, but only on non-multitextured surfaces (i.e. vertex lighting)