I could try my K5-PR133, possibly run it a bit higher. Not entirely sure that my GA-5AX supports it, if not, I'd have to use an older system.

I have a K5 PR166 running at PR200 on a VIA MVP3 board. Most of the games as Diablo 1 and Starcraft run already well on a fast 486, so the K5 runs them easily.
Still if you suggest benchmarks you should also mention the exact setup of the games and how you measure scores. Also the used graphics card is a factor.
So could you be a bit more specific?

Graphics card can be either S3 or Matrox, PCI obviously some shitty ISA/VLB would bring down the system greatly. As for scores, I guess FPS would do. Thanks for the quick answers by the way.

If anyone has a large selection of cards, I figure a PCI S3 Trio64+ and Matrox Millenium would be the most representative ones, for "regular computer" and "a bit more hi-end", respectively. To be honest, I never saw a 1995-1997 computer that had anything OTHER than a PCI S3 Trio64+ in it.

EDIT - If anyone has a Voodoo 1/2, then it would be nice to know the performance boost in 3D games as well.

K5 is interesting in that in synthetic tests its integer performance is similar to a Pentium Pro and faster than K6 on a per clock basis. But its FPU is more like that of a 486. In Quake it actually beats K6 per clock (i did a 133mhz comparison years ago). Unfortunately its inability to clock higher made it a dead end.

This is the 5k86 version, not the somewhat slower SSA/5.

How does it do in Quake in 320x200?

Just run Quake at 320x200 with sound on my PR133. Demo2 does 24.1fps.

m1so wrote:

How does it do in Quake in 320x200?

I don't have the results anymore. I remember the K5 @ 133 MHz (PR200) beat the K6 @ 133 MHz though. But the Pentium MMX / PPro / PII @ 133 are in another class. This isn't surprising with Quake though because the game is basically written for P5, and P6 is a monster compared to the other chips.

F2bnp wrote:

Just run Quake at 320x200 with sound on my PR133. Demo2 does 24.1fps.

Thanks. Quite nice for a chip that supposedly has a "486-like" FPU (and better than how Warcraft 3 plays on my netbook on which I'm typing right now). What FPS do you get if you turn the screen size down a bit? Nevertheless, it is a very nice result considering the real Pentium 133 gets around 30 fps (just 6 fps higher) if I am right. What graphics card are you using and how much RAM?

Do you have Duke Nukem 3D on the machine? Can you post the results for DNRATE on 320x200 and 640x480 with sound and enemies (with UNIVBE and DisplayDoctor [DisplayDoctor only if needed])? Again thank you very much.

Voodoo 3 AGP + 256MB RAM. This system usually houses a K6-III+ 550 so...

I have DN3D, although I only have the shareware version installed, can I still do DNRATE?

I think yes. It is just a cheat that shows your framerate. You just type it in in the middle of the game. Voodoo 3 and 256 MB RAM ... hmm very nice 😀 . How does GLQuake run in 640x480 (and lower if it can be set).

I just compared AMD K5 PR133 to Winchip C6 180 Mhz and ... surprise... the K5 wins in all performance parameters and as much as 2x in some categories, especially FPU, despite it being clocked at just 100 Mhz vs. the Winchip at 180 Mhz. The Winchip was a serious piece of steaming shit.

m1so wrote:

I just compared AMD K5 PR133 to Winchip C6 180 Mhz and ... surprise... the K5 wins in all performance parameters and as much as 2x in some categories, especially FPU, despite it being clocked at just 100 Mhz vs. the Winchip at 180 Mhz. The Winchip was a serious piece of steaming shit.

this is no surprising... the k5 actually has pretty good fpu, only a bit slower than pentium and comparable to k6-3 clock to clock. winchip on the other had the worst fpu, actual performance is about half of the pentium and synthetic tests are even lower, at one fourth of the pentium!

Well, I noticed the PR200 setting with the PR166 CPU runs unstable in certain situations. Duke3D seems to be a very good candidate which starts to report errors in GAME.CON on startup reliably.
However, I did run some benches I had on the disk:
System:

I also did run DOOM and got 90.1 fps.

Now clocked at PR166, 116 MHz:

And finally Doom gives 84.78 fps.

Nice one eliana. I was willing to do most of these myself but turns out you did them yourself! QuakeGL is especially CPU limited it turns out. At 100MHz it gives me around 17fps no matter what. Tried 640x480 and 512x384 and it produced the same framerate.
Also gave Jedi Knight a try, seems playable at 320x200 with my Voodoo 3, but there a lot of hiccups still, when with a faster CPU it should be a very smooth experience.

M1so, the K5 series was great. Clock for clock, it was a very efficient CPU, like swaaye said integer performance was better than the Pentiums. Unfortunately for AMD, it couldn't clock high enough and it was way too late to the market. The PR133 arrived in late 1996 to early 1997. The Pentium 133 like you said was out back in 1995. By the time AMD was ready to release the PR200, the K6 was almost out so they didn't give it an official release.

It's no wonder elianda couldn't get a fully stable 133MHz clock. I wouldn't be surprised if I couldn't run mine at 116.5 MHz which is the clock speed of the PR166.

m1so wrote:

I am aware of the Ultimate 686 test, but I am more interested in realistic game benchmarks rather than doing unplayable things like running Quake in 640x480 on it.

Thank you for this feedback. The reason for running Quake at 640x480 was due to the range of CPU frequencies tested in this study. The PII-333 OD, for example, yielded a very playable 27 fps in DOS Quake at this resolution. If I were to do it all over again, I'd probably add 320x200 as an additional test as well as Quake GL. A suitable graphics card to run 486 CPUs through PIII's would probably be the Rage 128 VR. This is the fastest graphics card I could find which ran OpenGL in Quake without issue.

F2bnp wrote:

It's no wonder elianda couldn't get a fully stable 133MHz clock. I wouldn't be surprised if I couldn't run mine at 116.5 MHz which is the clock speed of the PR166.

When running the 686 benchmark comparison, the fastest stable frequency I could ge the PR166 to was 120 MHz. At 125 MHz (83.3x1.5), some benchmarks starting failing. At 133 MHz, I think I could only complete a few DOS tests.

F2bnp wrote:

Nice one eliana. I was willing to do most of these myself but turns out you did them yourself! QuakeGL is especially CPU limited it turns out. At 100MHz it gives me around 17fps no matter what. Tried 640x480 and 512x384 and it produced the same framerate.
Also gave Jedi Knight a try, seems playable at 320x200 with my Voodoo 3, but there a lot of hiccups still, when with a faster CPU it should be a very smooth experience.

Thanks for all the feedback folks. It seems like a very nice CPU for the mid-1990s, with most games of the time running nice and smooth (except for GLQuake). But how come it gets less FPS in GLQuake than the software version? Didn't GLQuake get 30 fps even on Pentium 100? Are you running Jedi Knight accelerated or unaccelerated?

Also, elianda, did you run UNIVBE for this test? And can you try running Quake II in software mode in 320x200 please? What resolution did you run Quake II accelerated here?

Does anyone here have Descent 2 by any chance? FPS testing there would also be easy, just type in "frametimer", without brackets of course.

m1so wrote:

But how come it gets less FPS in GLQuake than the software version? Didn't GLQuake get 30 fps even on Pentium 100?

Anything higher than 320x200 probably kills the CPU, it just can't cope.

m1so wrote:

Are you running Jedi Knight accelerated or unaccelerated?

I tried both. I was talking about Accelerated at 320x200. Unaccelerated, the game needs to be played at around half the screen to start becoming playable, although I'd still consider it rather unplayable.

m1so wrote:

And can you try running Quake II in software mode in 320x200 please? What resolution did you run Quake II accelerated here?

I could do that.
I had Revenant (1999) installed as well. Semi playable indoors, I'm guessing it would have been awful on outdoor areas with monsters to fight.

What would be a good minimum configuration to run Revenant at least at 30 fps? I played it a lot on my 1 Ghz Coppermine/320 MB RAM/Geforce 2 MX400 and it obviously run well, but I am curious about how low can you go with this game while keeping it nice and smooth.

The official system requirements mention a 233MHz CPU. An MMX 233 should play the game quite nicely, although I'd guess a PII 300 would be enough to top the game.