foil_fresh wrote on 2021-09-06, 06:40:

i'd have never known there was any problem with the V3 running tomb raider if you never said this... to me, the v3 was running it flawlessly 🤣

I tried playing Tomb Raider on my Voodoo3 for a while, and those were the issues that stood out for me. More details:

• Lara's missing shadow - makes it a bit harder to drop down from high ledges (since you can't see where she'll land) but otherwise, mostly a cosmetic issue
• Pistols flashing with white outlines when used - this one is a bit more annoying, but it goes away if you by turn off anti aliasing
• Incorrect gamma - Adjusting gamma values under the Voodoo3 Windows driver panel had no effect on DOS Glide games. You can somewhat compensate for this by tweaking the brightness/contrast on your monitor, but that never looked right to me. OTOH, you can use these variables on the Voodoo1 to fine tune gamma as needed

Additionally, I don't think it's possible to control the refresh rate of DOS Glide games on a Voodoo3, so they always run as 60 Hz. On the Voodoo1, you can once again use environment variables such as SET SST_SCREENREFRESH=85 to adjust that. This is only relevant when playing on a CRT monitor, where 60 Hz produces way too much flicker. I don't remember if it was possible to override the refresh rate using VBEHz though.

I never was able to set the refreshrate on my voodo 2 either. Shadows are only there in software mode, voodoo 1 or ATI rage versions I believe. I also had very strange colour setting but I believe it was fixed with an environmental variable. The and aliasing and mipmapping functions seem to have a general problem.

So voodoo 2 and 3 are equally (bad) for tombraider 1, I guess. Reason to get a voodoo 1.. nah.

drosse1meyer wrote on 2021-09-06, 01:00:

voodoo3 on a p1 what is this madness haha

As always, it depends on what games you play and what resolution. There certainly are cases where a Voodoo 3 would be the much better option for a Pentium MMX build.
For example, an overclocked Pentium MMX 233 (running at 266 or above), will be bottlenecked by a Voodoo 2 in GLQuake even at 640 x 480 (the Voodoo 2 is capable of sustaining 108 FPS in GLQuake at this resolution, while the overclocked CPU can go up to ~ 130 FPS).
So a Voodoo 3 would be quite helpful in this case, because you could play at even higher resolutions, while also maintaining high framerates. Of course, Voodoo 2 SLI would also be an option, but I feel like these belong in a faster system. 😀
A Voodoo 3 is also very good for 1024 x 768 in Unreal. It works quite well with an overclocked Pentium MMX (35 - 40 FPS average). This would've been an extremely playable experience back in the day (I think most of us keep forgetting that "playability standards" have changed a lot since then. 😁

As for compatibility, my approach is to have multiple variants of the same platform (yes, I'm sick, I know it 😁 ), which is why I have:
- Pentium MMX 166 with Voodoo 1 (the most compatible card, but I wouldn't use it on a more powerful CPU)
- Pentium MMX 233 with Voodoo Banshee (it's a good fit, but will still bottleneck the CPU in many titles, even at very low resolutions)
- Pentium MMX 233 @ 266 MHz with Voodoo 2 (a good balance, fast while also maintaining compatibility)
- Pentium MMX 233 @ 292 MHz with Voodoo 3 2000 (this build surprised me a lot, it's actually extremely fast for what it is - this would've been mind-blowing to me back in the day)

renejr902 wrote on 2021-09-06, 05:28:

I bought a voodoo 3 because i had a better deal than for a voodoo 2. And a voodoo 3 can be use in my pentium 2 and 3 too with great performance.

No problem there considering the prices today, the reason why I said Voodoo2 was for better compatibility.

Joakim wrote on 2021-09-06, 07:23:

I never was able to set the refreshrate on my voodo 2 either. Shadows are only there in software mode, voodoo 1 or ATI rage versions I believe. I also had very strange colour setting but I believe it was fixed with an environmental variable. The and aliasing and mipmapping functions seem to have a general problem.

So voodoo 2 and 3 are equally (bad) for tombraider 1, I guess. Reason to get a voodoo 1.. nah.

Voodoo2 uses slightly different environments variables for setting the refresh rate (there are even for each screen resolution) on DOS.

1SSTV2_SCREENREFRESH=60
2REM The following variables can be used if SSTV2_SCREENREFRESH is not set
3SSTV2_REFRESH_512x384=60
4SSTV2_REFRESH_640x480=60
5SSTV2_REFRESH_800x600=60
6SSTV2_REFRESH_1024x768=60

There are different versions of Tomb Raider for 3Dfx which one can run on later cards but lack the shadows (Voodoo Rush patch).

At least on Voodoo2 you can use the environment variables that are found on V2-AUTO.INF file from the drivers that allows you to run executables meant for Voodoo Graphics (especially those that uses GLIDE.DLL or are statically linked with the Glide API), it's not perfect but will run the good titles.

renejr902 wrote on 2021-09-06, 05:28:

I bought a voodoo 3 because i had a better deal than for a voodoo 2. And a voodoo 3 can be use in my pentium 2 and 3 too with great performance.

Of course, I wish I could/did find a v3 PCI for less than a v1/v2 😉

What board do you have? it seems both my socket 7 mobos don't like to give video with newer (1999+) PCI video cards, wonder if the V3 would be the same...

bloodem wrote on 2021-09-06, 07:43:

As always, it depends on what games you play and what resolution. There certainly are cases where a Voodoo 3 would be the much better option for a Pentium MMX build.
For example, an overclocked Pentium MMX 233 (running at 266 or above), will be bottlenecked by a Voodoo 2 in GLQuake even at 640 x 480 (the Voodoo 2 is capable of sustaining 108 FPS in GLQuake at this resolution, while the overclocked CPU can go up to ~ 130 FPS).

The voodoo3 is basically an overclocked V2 with more memory and 2d added in, so that makes sense. As far as bottlenecks, according to Phil's PDF, it looks like for example with quake2, the CPU is the bottleneck with V2s even with SLI, unless I am misunderstanding his results. I would be interested to see more info about your Unreal test setup as it seems he gets mid 20s @ 640x480 with a V2 on P233.

My voodoo1 on P233 gets barely 20 fps in opening flyby in 640x480 🙁 Though it is playable in game roughly 20-30fps, it's always that damn timedemo which kills my overall FPS report.

drosse1meyer wrote on 2021-09-06, 13:28:

The voodoo3 is basically an overclocked V2 with more memory and 2d added in, so that makes sense.

Actually it's a Voodoo Banshee with dual TMU with higher clocks and lacks the Voodoo2 registers.

Gmlb256 wrote on 2021-09-06, 13:59:

drosse1meyer wrote on 2021-09-06, 13:28:

The voodoo3 is basically an overclocked V2 with more memory and 2d added in, so that makes sense.

Actually it's a Voodoo Banshee with dual TMU with higher clocks and lacks the Voodoo2 registers.

I thought the banshee lost the second TMU which the voodoo2 originally had?

drosse1meyer wrote on 2021-09-06, 13:28:

As far as bottlenecks, according to Phil's PDF, it looks like for example with quake2, the CPU is the bottleneck with V2s even with SLI, unless I am misunderstanding his results. I would be interested to see more info about your Unreal test setup as it seems he gets mid 20s @ 640x480 with a V2 on P233.

Quake 2 is a different kind of animal, that one is indeed CPU bottlenecked when using a Voodoo 2 on a Pentium MMX. GLQuake, on the other hand, is not.
A Pentium MMX 233 gets a substantial boost when overclocked to 266 and especially to 292 MHz (FSB 83) on motherboards that support it. So, yeah, Unreal becomes very playable on such a system.

drosse1meyer wrote on 2021-09-06, 13:28:

My voodoo1 on P233 gets barely 20 fps in opening flyby in 640x480 🙁 Though it is playable in game roughly 20-30fps, it's always that damn timedemo which kills my overall FPS report.

The Voodoo 1 is a really slow card, it will not be able to deliver decently playable framerates at 640 x 480 in Unreal, even if CPU bottleneck is completely eliminated. That being said, it's still a nice card for what it is and it's a great pair for a slower Pentium MMX (like the 166).

drosse1meyer wrote on 2021-09-06, 14:08:

I thought the banshee lost the second TMU which the voodoo2 originally had?

Internally, "Banshee" was the code name that 3dfx used for the new Voodoo 3. Eventually, Banshee became the name of their low cost card, targeted at OEMs.
All in all, as Gmlb256 said, the Banshee is much more similar to the Voodoo 3, than it is to the Voodoo 2.

it's actually been "Banshee 2" in development. The Voodoo3 name is just STB things, and a lot of the press went on like it's literally "Voodoo2 SLI one card" when that's only happened with certain Obsidian products. Banshee had that same "22-bit" filter too.

According to Scott Sellers, internally, during development, they simply called it Banshee.
I would guess that the general public did call it Banshee 2. 😀

BitWrangler wrote on 2021-09-05, 22:04:

The datasheet would make a point if it wasn't dated June 97, i.e. after all the P54 multipliers had been introduced and some 3+ years since the beginning of the family.

The datasheet provides information about all of these CPUs (from 75MHz to 200MHz, but excluding the Socket 4 ones which did not come with any CPU multipliers afaik), which includes the ones released earlier. But at least I provided a solid foundation for my info, which you have not.
I'm not sure why you are having such a hard time accepting this.

What info??? All that says is "as of now June 1997, this is the function of the pins" I cannot see a single sentence in the thing which makes you think you've proved your point.

edit: Here's some actual meat and potatoes, http://bitsavers.trailing-edge.com/components … _Components.pdf page 113, see that, not a BFO and BF1 in sight, just a BF pin that's tri state, if they were locking out higher multis at this point in time they would have planned better, they weren't because they didn't exist yet.

drosse1meyer wrote on 2021-09-06, 13:28:

Of course, I wish I could/did find a v3 PCI for less than a v1/v2 ;) […]
Show full quote

renejr902 wrote on 2021-09-06, 05:28:

I bought a voodoo 3 because i had a better deal than for a voodoo 2. And a voodoo 3 can be use in my pentium 2 and 3 too with great performance.

Of course, I wish I could/did find a v3 PCI for less than a v1/v2 😉

What board do you have? it seems both my socket 7 mobos don't like to give video with newer (1999+) PCI video cards, wonder if the V3 would be the same...

bloodem wrote on 2021-09-06, 07:43:

As always, it depends on what games you play and what resolution. There certainly are cases where a Voodoo 3 would be the much better option for a Pentium MMX build.
For example, an overclocked Pentium MMX 233 (running at 266 or above), will be bottlenecked by a Voodoo 2 in GLQuake even at 640 x 480 (the Voodoo 2 is capable of sustaining 108 FPS in GLQuake at this resolution, while the overclocked CPU can go up to ~ 130 FPS).

The voodoo3 is basically an overclocked V2 with more memory and 2d added in, so that makes sense. As far as bottlenecks, according to Phil's PDF, it looks like for example with quake2, the CPU is the bottleneck with V2s even with SLI, unless I am misunderstanding his results. I would be interested to see more info about your Unreal test setup as it seems he gets mid 20s @ 640x480 with a V2 on P233.

My voodoo1 on P233 gets barely 20 fps in opening flyby in 640x480 🙁 Though it is playable in game roughly 20-30fps, it's always that damn timedemo which kills my overall FPS report.

On this 486 motherboard with pci:

Luckystar ls486e rev c1

the pci voodoo3 runs perfect, no problem at all, works great on dos and glide games and windows 95 too. But with the pci voodoo banshee, it works great on dos and glide, but i cant get any windows 95 banshee driver to work with it, i tried for a long time, i tried every driver version existing.

I have no problem with my Pentium 233 mmx and my Pentium 2 and 3 to get the pci voodoo 3 working. I have 2 pentium 2 and 3 pentium 3 and the voodoo 3 works perfect with both of them. If you want more motherboard information, i can tell you more, no problem. 😀

bloodem wrote on 2021-08-18, 17:32:

luk1999 wrote on 2021-08-18, 16:42:

@drosse1meyer
You can always try to overclock your cpu to 200 MHz (just set multiplier to 3x) and check how much your framerate improve.

It depends on the CPU. Some of them are multiplier locked (limited). I have a Pentium 166 (non-MMX) which will ignore multipliers higher than 2.5 (basically, it will just ignore the 3.0 multi, because 3.5 is interpreted as 1.5 on non-MMX CPUs).
As for Pentium MMX, I think they're all locked (limited) except for the 233 SKUs (which go all the way up to 3.5). The others only accept lower multipliers.

Hi! Bloodem I hope you're doing great ! You was the first one that answered me in the vogons forum when i was a newbie 1-2 years ago and you helped me a lot.

Your post just remember me that my pci voodoo 3 is in my pentium 233 mmx ( not in my pentium 200 mhz mmx ) but i'm not sure it would do a big difference for this guy on his pentium 200 mmx.

BitWrangler wrote on 2021-09-07, 01:07:

What info??? All that says is "as of now June 1997, this is the function of the pins" I cannot see a single sentence in the thing which makes you think you've proved your point.

edit: Here's some actual meat and potatoes, http://bitsavers.trailing-edge.com/components … _Components.pdf page 113, see that, not a BFO and BF1 in sight, just a BF pin that's tri state, if they were locking out higher multis at this point in time they would have planned better, they weren't because they didn't exist yet.

Cheers. I checked and it looks like you were indeed correct regarding the BF pin on the oldest s5 Pentiums!
I never even knew this. This means that what you said here

Early 75, 90, 100 probably only had 1.5 implemented too, but CC0 stepping upward all seem to do 2x as well.

is probably also correct.
Interesting..

It would still be best to actually test this though.

Btw, I do think they did plan ahead, or at least to a certain degree. In your link on page 13 you can see the BF pin (pin 33Y) with pins 34X and 35W set as NC (Not Connected), where according to this on page 8 the pins 33Y and 34X are BF0 and BF1 respectively, with 35W still NC. Apparently they did leave room for future 'upgrades' but ended up never using 35W for a BF pin.

Still kinda odd for Intel to produce a specsheet saying all these CPUs had both the BF pins connected, but my guess is that all of them did have a BF0 and BF1 at that time (or else there wouldn't have been the need to engrave the i133 into the bottom of some of the Pentium 133MHz CPUs if these didn't support any higher CPU multipliers than 2x.
And Intel was probably not selling the oldest chips anymore at the time, so they would have had no quarrel removing this information from their 1997 specsheet.

But again, would be best to try it out for real (will be tough getting a suitable motherboard for this task though as most boards didn't go below 50MHz FSB anyway and the earliest Pentiums probably didn't overclock very well).

Tetrium wrote on 2021-09-07, 14:45:

8><CUT
Still kinda odd for Intel to produce a specsheet saying all these CPUs had both the BF pins connected, but
8><CUT

Over the last couple of years I have tried to research certain stuff using Intel specsheets and I can almost assure you that it is not odd at all for Intel specsheets to contain errors and/or to be not 100% complete on certain aspects. Considering that errors occured in the hardware it can be expected that errors also occur in the documentation - probably especially(?) in versions that were updated in later years.

renejr902 wrote on 2021-09-07, 02:52:

Hi! Bloodem I hope you're doing great ! You was the first one that answered me in the vogons forum when i was a newbie 1-2 years ago and you helped me a lot.

Your post just remember me that my pci voodoo 3 is in my pentium 233 mmx ( not in my pentium 200 mhz mmx ) but i'm not sure it would do a big difference for this guy on his pentium 200 mmx.

Hi! Yeah, I remember! Glad to have been of help. 😀
Doing good (great is too big a word 😁), thank you!

PARKE wrote on 2021-09-07, 15:06:

Tetrium wrote on 2021-09-07, 14:45:

8><CUT
Still kinda odd for Intel to produce a specsheet saying all these CPUs had both the BF pins connected, but
8><CUT

Over the last couple of years I have tried to research certain stuff using Intel specsheets and I can almost assure you that it is not odd at all for Intel specsheets to contain errors and/or to be not 100% complete on certain aspects. Considering that errors occured in the hardware it can be expected that errors also occur in the documentation - probably especially(?) in versions that were updated in later years.

Very true.
This is basically why I believe that real world experience is always better than technical datasheets as the paper can still be wrong regardless.

Tetrium wrote on 2021-09-07, 15:49:

This is basically why I believe that real world experience is always better than technical datasheets as the paper can still be wrong regardless.

People have, I believe, been hunting down 1500 MHz Tualatin Celerons for a number of years because two of them were listed on page 26 of the Intel®Celeron® Processor Specification Update - aug 2008 (24374853.pdf).
https://www.manualowl.com/m/Intel/SL3VS/Manua … /362392?page=18
Just an/one example...

bloodem wrote on 2021-09-06, 18:48:

According to Scott Sellers, internally, during development, they simply called it Banshee.
I would guess that the general public did call it Banshee 2. 😀

Yeah i've heard of a Banshee 2 before there was the Voodoo3 name at one point... 🤔