I know about that multi-texturing advantage. In fact 2 textures per pass, since this multitexturing word is a bit too general. The massive performance boost: probably in SLI mode. Otherwise the boost is not so big.

I remember seeing a Savage 4 (110 GPU/ 125 Memory) running with and without Multitex, with and without texture compression (Quake3 in 16 bit color max details). The card was having an advantage on 1024x768 for texture compression (3-4fps more) otherwise the advantage is lower at 800 or 640 resolutions.

Regarding multi-texturing, the gain was almost similar in 1024 resolution as when the texture compression being used (3-4 fps more). In 640 and 800 the difference was less visible. Even without those two options, the card was still able to do more than 20 fps in 1024x768x16.

And the Banshee (110 GPU/ 125 Memory) was able to do a bit less than Savage 4 with multitexturing (no texture compression), but still above 20 fps. Multitexturing benefits varies upon situation, but it gives a certain 2-4 fps increase in most situations.

Quality is more of a concern for non-single pass multitexturing cards mostly when you need bump-mapping (emboss most of the time). Then, a non-Mtex card is showing less quality, even when it is forced to do the effect as precise as possible. But other than Slave Zero and few other games, bump-mapping is kind of rare for 3D games. This is why non-Mtex is actually not such a big loss in quality most of the time, but a certain loss in speed (gets bigger as more texture are required for an object).

I admit that the comparison might not be the best with Savage 4 and Banshee. But Savage with and without multi-texture is at least a fair proof of what you are most likely to get, and it is not so much. A possible response to this question is that you can never forget that the CPU and PCI bus are limiting your performance at first, and the actual software code is the biggest difference sometimes. And finally there are games and games. Quake is not a good example of multi-tex. Apart from some shadowmaps and some lightmaps, the whole game is kind of single textured objects. Sometimes they never use Mtex, other times they use it heavily (Unreal Tournament).

Quake 2 is a different problem. First of all Quake 2 normally used a different path for the OpenGL wrapper (had a special one for a Quake 1 and a special one for Quake 2 as opposed to the only full OpenGL ICD available for Banshee). Remember the fxmemmap.vxd file in Quake2 directory? Not to be forgotten is that on a Banshee the GPU must also refresh the 2D display frames at the same time, while the Voodoo2 lets this job on the 2D card. The speed hit can be calculated on paper, and it is not indiferent for the performance.

Also, a test on Quake 3 with an OpenGL for a Voodoo2 (and Quake3 only) proved that the Banshee could get a few more fps. So again, comparing Banshee and Voodoo2 and solely saying that a big performance is gained by the usage of Mtex is not founded.

Yes and that is only with a Voodoo2 consider what to voodoo2's can do!

PS. if the voodoo2 is used in SLI. Does it than use 2 PCI busses ( I thought so). Or does all the DATA goed trough the SLI cable. Probably not, but maybe I am wrong. But also a banshee only uses AGP1 64bit 66mhz.
If you have 2 PCI slots with 64bit and 66mhz you have the same bandwith as a banshee!!!!!!
The Banshee can't compete with that.
And yes most V2 SLI users use them allongside fast Geforces/Radeons/V5's, and those PC's have (like I do) 2ghz+.

Falcongly (why you spell my name wrong?) I have to add the following: I am comparing a Voodoo2 (90/90)with a GA-630 Voodoo Banshee (110 Mhz GPU, 125 Mhz Memory).

I haven't figured out what test results you showed in here. It was a good thing to add the link to the test. The first Banshees are normally running at 100Mhz and have 8Mb video memory. Voodoo2 runs at 90Mhz and has 12Mb video memory. The fastest Banshees ran at 125Mhz and have 16Mb video memory. These improvements make a big difference, and the whole test with Quake2 is almost sensless. Quake 2 was playable (above 25fps on a Voodoo Banshee in 1024x768 or around 30 fps in 800x600 with no 3DNow! Quake). Voodoo2 normal was somewhere around 40-50fps. Actually not such a big difference.

I would refrain on posting this kind of test results like you did, without saying what were the full hardware specifications of the Video card at least. And, also, a crappy motherboard can totally screw up a test. The Intel BX400 (I guess) was a very bad P2 motherboard, at least from a certain manufacturer: DTK.

And when you think that you play a game usually at 1024x768 (when it performs acceptable), the Banshee becomes more than attractive compared to a normal Voodoo2. And the 16Mb texture memory start to make a difference in games better than Quake2, even if the card gets a slight performance hit when using multi-pass single-texturing. Most of the time is pretty useless what performance you get in 640x480 resolution, since most of the time you wouldn't use the card in that resolution.

Always the Banshee was put at tests against Voodoo2 in SLI and in not so great games such as Quake2. Quake3 really pushed the limits of the Voodoo2, even in SLI. If the game is not stressing too much the cards, you don't see some of the benefits of the Banshee. Also, when Banshee can usually outrun a Savage4 with single-pass multitexturing running at the same frequency, you draw the conclusion that Voodoo Banshee is a very good card, and in some respects clearly better than the Voodoo2.

Quake2. The first thing that comes to my memory is the special Openg32.dll released by ID (with 3dfx support). This is not a full OpenGL ICD. Quake 1 had this kind of OpenGL to Glide wrapper. These so-called miniGL had been long time optimised for Voodoo1 and Voodoo2. For Voodoo Banshee, 3dfx was working on a full ICD at that time. I remember the poor fix for NFS3. It looked horrible in Glide when this patch was applied. What I want to point is that when Banshee appeared, a long time it didn't had the best support as oposed to Voodoo2. Especially games as Quake2 that were designed from the start for Voodoo2.

Full OpenGL ICD. I remember seeing a Banshee running with the latest Quake 3 OpenGL for Voodoo2. Apart from the problem of multi-texturing, there was a certain improvement, sometimes of more than 5 fps in 800x600 resolution, as oposed to Banshee full OpenGL ICD. And don't forget miniGL is based on glide2x (smaller and faster for simple jobs), the full ICD is based on glide3x. This can make for some subtile differences. Testing both cards in Glide2x on a game would be the best compromise in seeing how they handle the load.

I have pointed in one direction. In certain aspects Banshee is better than Voodoo2. Even if it lack the 2 textures per pass, the real situations does not depend so much on this. Ok, you get 10, 15 fps. But again, using a low resolution. Also, to be noted is the fact that quality is lower on a Voodoo2 as oppsed to a Banshee.

The whole point about multitexturing and it's impact on performance was this: you certainly get a fairly good improvement, depending on the situation, but it is not solely the reason for which the Voodoo2 could outperform the Banshee. Until Voodoo3 arived there was no card to beat up the performance of a Voodoo2 in SLI. And this clearly proves some excelent performance capabilities from the Voodoo2.

And getting a whole discussion back, a Voodoo2 would not offer too much of an improvement if the GPU would be overclocked, unless the memory response would be also fasten up.

I know there is one clock for GPU and Memory on Voodoo2. I said this earlier. The whole problem was what would happen if the GPU would be clocked faster as opposed to what would happen if the Memory would be faster. Pure theory...

Can you do that clock change for any layer? D3d/OpenGL/Glide. I might be wrong, but I remember that only in Glide/OpenGL you can do that. Sorry if I am wrong. I know little about this settings effect.