Very nice benchmarking. Might be interesting to compare results to that of the '686 comparison' sticky, but different hardware configurations might already make direct comparisons troublesome. In it the Pro and mmx seem more evenly matched in for example Quake 2.

Good to see the pro keeping up... except in Doom. Doom really doesn't care about the pro. Maybe a lot of optimized 16 bit register accesses going on in there that trouble the original P6?
Anyway, very nicely done.

amadeus777999 wrote:

auron wrote:

some interesting results there. i wonder if the pentium pro suffers any penalty from running win98se (mixed 16/32-bit code), it would be interesting to rerun some of this on win2k. of course the pii has better l1 cache as well, i believe.

mmx seems nearly irrelevant in most cases from what i can tell, unless diablo 2 actually uses it, but i'd take the "pmmx>ppro" conclusion with a grain of salt for now. generally i think it would have been better to do all tests with the same amount of memory, 64mb to not penalize the socket7 system and use the same hdd for all.

What was the general loss in performance when running 16/32 bit code on the PPro? I can't imagine this being as vital as was propagated - seems more like a maketing "gag" of that era.

This would be a nice continuation test for these three systems. Running a program with heavy 16 bit register ussage, first on the original pentium/pentium with mmx, then on the Pro that had trouble combining its OOO execution with partial registers and finally the Pentium II that supposedly fixed this.

There are likely not that many 16 bit register accesses - maybe when selecting the planes or in the code of the DMX libraries.
There's unfortunately no original Doom source-code, only the Kreimeier version, but I strongly doubt that the 16/32bit dilemma is as dramatic as assumed. There is of course a part of the OS, be it 9x and/or DOS, which has parts that execute 16 bit code that can't of course be circumvented by using a DosExtender... one is bound to suffer on the underlying system.
Nonetheless the PPro is less than 4% slower than the PII here.

auron wrote:

mmx seems nearly irrelevant in most cases from what i can tell, unless diablo 2 actually uses it, but i'd take the "pmmx>ppro" conclusion with a grain of salt for now. generally i think it would have been better to do all tests with the same amount of memory, 64mb to not penalize the socket7 system and use the same hdd for all.

when you say "pmmx>ppro" do you mean in the case of Diablo 2? If so, I can tell you that this was the case, but I just couldn't substantiate it unfortunately, which is why I took rough notes. I had a few other games tested, such as Gabriel Knight 3, Z.A.R., Carmageddon 1&2, Revenant, Dungeon Keeper, Quest for Glory V, Tex Murphy Overseer and quite a few more.
The problem is that with most of these there's not really an easy way to test performance and I just had to draw the line somewhere, so I took some rough notes when trying these out just for curiosity.

I picked Jazz 2 and Diablo 2 only because they were very interesting, otherwise I wouldn't have included them.

As far as taking it with a grain of salt, suit yourself, it is what it is. The Pentium Pro was chugging hard on my end, where as the Pentium MMX was quite nice and acceptable. If you're asking me about playing the game nowadays, I wouldn't have played it on either of the 3 systems I tested, just because I can run it on more modern hardware and ensure that I never get a single stutter or framerate drop.
Then you say that I shouldn't have penalized the socket 7 system with 64MB RAM, but it comes out on top in Diablo 2 which you use as your example, so...
It might have been fair to include 64MB tests on the side, but it would have cluttered the results even more and more importantly, I was too fucking bored to pull it off. You are free to try of course. As far as the HDD goes, all systems were using "modern" HDDs by their standards, 20+GB disks made well into the 00's. I think UDMA 33 was more than saturated there and 440FX doesn't even feature that so maybe there's your problem right there.

didn't mean to doubt that you wrote down what you saw, it's just when trying to get a picture of pure cpu performance (which is the point of this test, from my understanding) other factors should be excluded as much as possible ideally. if you've used fast hdd's on all systems that's fair enough, you didn't mention this before. good point about the dma limitation on 440fx but why would the 96mb system with nominally faster cpu access the hdd more than the 64mb system with weaker cpu (that's how i understood your comments)? does the game use ram inefficiently? just trying to make sense of the results here.

as for speculation why this game could run better on pmmx vs ppro, maybe it uses the 32kb l1 on the mmx chip well enough that there is a penalty on ppro, mmx instructions, or another factor. turok 2 seems another one rare case where the ppro and pmmx are awfully close, with the pii pulling ahead by roughly 20%, but with the pii system having used more and faster ram it's hard to tell how much of that is due to the cpu itself.

Good work. Most results are "as expected" (II > Pro > MMX), but I'm curious what happened in DOOM. AFAIK it was compiled with the 32-bit Watcom compiler so it PPro shouldn't have a problem with 16-bit code.

GL1zdA wrote:

Good work. Most results are "as expected" (II > Pro > MMX), but I'm curious what happened in DOOM. AFAIK it was compiled with the 32-bit Watcom compiler so it PPro shouldn't have a problem with 16-bit code.

Pentium II is also very very close to the MMX here, so I don't know whats' going on either.

GL1zdA wrote:

Good work. Most results are "as expected" (II > Pro > MMX), but I'm curious what happened in DOOM. AFAIK it was compiled with the 32-bit Watcom compiler so it PPro shouldn't have a problem with 16-bit code.

The Extender is still running on top of Dos. For some calls it has to switch between "worlds" I assume.

Doom is an old game and therefore the fpu is not used. But at over 80 fps it doesn't matter that the mmx is ahead of P Pro. More important is that the Pentium Pro has a much more powerfull Fpu unit that Pentium MMX has and this is counting in the more intensive games like Quake and Unreal engine based games.

Interesting results.

Would be interesting if a 440LX for the PII is also ahead the others.
The LX chipset is the more time correct chipset, cause the BX release was with the PII-350 with 100MHz FSB variant.

We don't talk about old threads. Once they've moved to page 5 of the results, they are dead to us. Can't nothing be relevant about year old conversations about 20 year old computers.

So say we all.

-- Internet forum etiquette guidelines.

(I keed. 😁)

I ran a similar comparison to this in another thread. Thought they should be linked in case one stumbles upon the other. Re: NEC Proserva V Plus (dual socket 7 system) (have a lead on multiple of these)