Depends on your wallet and availability. Voodoo 1 is more cheap option (usually), but performance is not optimal for a Pentium Pro system.

Voodoo 1 would be a good match, Voodoo 2 too, although the CPU would be a big limit there. Anything over that would be overkill (and would replace the Millennium II as they are all standalone cards)..

How is the output quality between Voodoo 1 and 2? The matrox was known for great quality and I'm curious how it's impacted when passing the signal through the voodoo. How is the quality affected when the voodoo engine is not engaged vs when engaged in supported games?

For a PPro 200, only a Voodoo1 makes sense, really. A Voodoo2 is a bit wasted, as it needs a CPU with at least twice the power to show what it can do.

On the other hand, a Voodoo2 has generally better 3d graphics quality, can do 800x600 with z-buffer, and has more texture memory. I'd say use what you can get cheap. But if you want to do it period correct, you should go for a Voodoo1.

Pentium Pro 200 performance is more or less equal to Pentium II 200. Voodoo 2 is fine for such setup.

Voodoo Graphics (Voodoo1) is the most "period correct", but also a good match regarding performance. Voodoo2 also works but it appeared much later, also it will be much more limited by the CPU. V2 starts shinning on PII 300 and above.

So I'd go with Voodoo1. But there's one more thing - Pentium Pro was not designed for games, and games weren't really designed for P Pro. In most games Pentium 200 MMX will probably run better, because games of that era were optimised for P5 (not P6), and a lot had patches for MMX. Plus a lot of games had 16 bit code, P Pro is much slower for 16 bit code, it's optimised for 32 bit code.

In many cases, Voodoo 1 would be too weak for Pentium Pro. Voodoo 1 fall short in Quake 2, Incoming, Shogo, etc.
And yes, Pentium Pro 256Kb = equally clocked Pentium II with PCI 3D accelerator.

Yeah, this has been proven time and time again. Here's another comparison:
Pentium Pro 233 vs Pentium MMX 233 vs Pentium II 233

Voodoo 2 is also much faster than a Voodoo 1 and because it offloads triangle setup among other things, the gains are apparent even on older and slower CPUs.

The Serpent Rider wrote on 2020-02-05, 09:09:

Pentium Pro 200 performance is more or less equal to Pentium II 200. Voodoo 2 is fine for such setup.

PPro is clock-for-clock faster than PII due primarily to the full speed L2 cache

MMX was 90% hype and didn't really mean shit

maxtherabbit wrote on 2020-02-05, 20:47:

The Serpent Rider wrote on 2020-02-05, 09:09:

Pentium Pro 200 performance is more or less equal to Pentium II 200. Voodoo 2 is fine for such setup.

PPro is clock-for-clock faster than PII due primarily to the full speed L2 cache

MMX was 90% hype and didn't really mean shit

This is nonsense, as performance between the P2 and Pentium Pro is largely going to depend on what software you are running.

Use 32bit software, the Pentium Pro will pull ahead slightly.
Use 16bit software, like was very common when the Pentium Pro was introduced, and the Pentium 2 will be quicker.

The differences between the 2 are not going to be earth shattering clock for clock, so I do not understand where all the angst is coming from.

I'll just add some pretty pictures of my own tests then to try and settle this:

Quake is from Phils Dosbench suite, fastvid or Univbe where appropriate.
Quake 2 runs under Windows 98SE.
Lightwave 3d 5.6 "Mustang render" is a render of one frame from a provided scene and ran under Windows 2000.

Pentium/MMX ran on an ASUS T2P4 with 256MB EDO memory.
Pentium Pro's on an Intel VS440FX also with the same 256MB EDO as above.
Pentium 2's ran on an ASUS P2B-LS with 256MB SDRAM.
The same Matrox Millenium on all of the systems.

Note that the PPRO's are running at 233MHZ, brainfart from my end.
But even at 233MHz the PPRO's cannot compete with the 266MHz P2.
Tho theres probably some benches that can ulilize the faster cache on the PPRO's.

bofh.fromhell wrote on 2020-02-06, 05:52:

I'll just add some pretty pictures of my own tests then to try and settle this:

seeing as how you never actually compared them clock for clock, I don't see the relevance

maxtherabbit wrote on 2020-02-06, 15:16:

bofh.fromhell wrote on 2020-02-06, 05:52:

I'll just add some pretty pictures of my own tests then to try and settle this:

seeing as how you never actually compared them clock for clock, I don't see the relevance

PPRO's never exceeded 200MHz and Pentium 2's started at 233MHz.
Any other frequency's would just be guessing.
You might do really good guesses, but its still guessing.

Clock for clock comparisons are usually irrelevant because the whole purpose of a "new" architecture is to allow much higher frequency's.
And thus easilly offsetting any IPC losses.

Sorry TS for the thread derailing!

bofh.fromhell wrote on 2020-02-06, 15:35:

PPRO's never exceeded 200MHz and Pentium 2's started at 233MHz.

you ran the Pros at 233 in your test... comparing them at 233 would be easy since just about every PPro would easily OC to 233

bofh.fromhell wrote on 2020-02-06, 15:35:

[....]

Clock for clock comparisons are usually irrelevant because the whole purpose of a "new" architecture is to allow much higher frequency's.
And thus easilly offsetting any IPC losses.

"Usually" - yes. But not in this case. The 32b core of the P2 is for all intents and purposes identical to the 32b core of the PPro, with performance differences being due to clocking and cache differences (256kB full speed usually beating 512kB half speed); the big difference in the P2 is that it is far stronger in 16b code, where a P2 beats a similarly clocked P55C, whereas a PPro performs far worse. This is one instance where a 3-way comparison between P55C, PPro and P2 at 233MHz makes a lot of sense.

maxtherabbit wrote on 2020-02-05, 20:47:

PPro is clock-for-clock faster than PII due primarily to the full speed L2 cache

That's debatable. In the application uses 32 bit code and doesn't use MMX then, yes, PPro is likely faster.
But if the application uses MMX and also rely on some 16 bit code, then PPro will be slower, in some cases a lot slower.
Also remember that PII has twice the cache of PPro (there are PPros with 512k and even 1 mb cache, but we're talking about the standard 256k version).

maxtherabbit wrote on 2020-02-05, 20:47:

MMX was 90% hype and didn't really mean shit

90% hype? Again, highly debatable. Yes, some apps do not use MMX at all. But some do use MMX and they do it very good.
I might be wrong, but I remember that Unreal Tournament was optimised for MMX, a friend of mine upgraded from P166 to P200 MMX, the difference in UT was huge.
Sure, it was software mode, and there was more than MMX, it was also a nice boost in cpu clock, but I'm sure MMX was a big factor in the speed boost.

In the end it all depends in what is the role of the machine. I repeat myself - if you want it to be as close as possible to "period correct" then V1 is the way to go.
If you want to play newer games then go for V2. But in this case there's no reason to go with PPro, you should go for PII or even PIII.

Personally, I'd never build a machine for games with PPro. PPro was an awesome CPU, but it was targeted for workstations and servers. Sure, it can play games, but it's not a good choice.
It's like going now for a low clock, but very high core count machine to play current games. You can buy a Xeon Platinum 9282 and pair it with RTX 2080Ti. Will it play games. Sure, it will, and very good.
But will it play it better than a 9900kf? No. It will only cost a lot more.

Are there some examples of the "much slower" 16-bit code performance? So far I have never seen PPro to perform worse than Pentium MMX at the same clock. Not in DOS and definitely not in any Windows SW or games. I had builds with 233 MHz PPro and 233 MHz MMX... and both performed about the same in everything 16-bit I tried.

And here is an example of some 3d accelerated games... clear win for the PPro every time http://hw-museum.cz/article/5/cpu-history-tou … 1995---1999-/12

Oh and I believe MMX was in fact just a hype. Most of the extra performance is there because of the doubled L1 cache of P MMX vs Pentium classic.

havli wrote on 2020-02-07, 22:55:

Are there some examples of the "much slower" 16-bit code performance? So far I have never seen PPro to perform worse than Pentium MMX at the same clock. Not in DOS and definitely not in any Windows SW or games. I had builds with 233 MHz PPro and 233 MHz MMX... and both performed about the same in everything 16-bit I tried.

https://www.tomshardware.com/reviews/intel-pe … um-ii,20-2.html

Head-to-head of P2, PPro and MMX (inexplicably clocked at 225MHz). MMX frequently beats PPro. Not massively, but given its cache runs at 1/3 of CPU speed vs full speed cache in PPro, that's an indication things are not good.

And here is an example of some 3d accelerated games... clear win for the PPro every time http://hw-museum.cz/article/5/cpu-history-tou … 1995---1999-/12

Yes, in 32b code the PPro shines.

Oh and I believe MMX was in fact just a hype. Most of the extra performance is there because of the doubled L1 cache of P MMX vs Pentium classic.

Of course. By the time MMX was actually used, we were in P3 era. The only MMX-era application where MMX was actually used was in WinModems...

But 32kB L1 cache on isn't to be sniffed at. It reduced the dependency on slow external cache and very significantly boosted performance. Those MMX instructions are almost irrelevant, but the cache maketh the CPU.

Not massively, but given its cache runs at 1/3 of CPU speed vs full speed cache in PPro, that's an indication things are not good.

That doesn't indicate anything. Both PPro and PII don't show significant boost in DOS games, compared to equally clocked PMMX.