VOGONS


Pentium II/III VS K6-III+

Topic actions

First post, by F2bnp

User metadata
Rank l33t
Rank
l33t

I have promised this thread for a couple of months now and so here it is (finally!). The aim of this research was to pinpoint the exact performance of a K6-III+ running at 550MHz and 600MHz relative to the Pentium II and III (Katmai) CPUs. Various CPUs were tested for this reason. I hope these tests will be helpful in understanding the K6 CPUs' performance a little better.

What did we test? Games of course! 🤣 I tried to make this somewhat diverse and include some less conventional games. I had to leave a few out since they had no timedemos or any other way of accurately measuring performance and so I had to rely on my own assessments.

I'm finally done putting this together! I did my best to make the charts easily readable, but this is my first time trying this, so feel free to criticize and comment on how I can make it better next time!

Special thanks to Imperious for providing MVP3 results and Keropi for repairing my MVP3 board!

Attachments

  • Filename
    Benchmarks.zip
    File size
    209.01 KiB
    Downloads
    359 downloads
    File license
    Fair use/fair dealing exception
Last edited by F2bnp on 2018-05-03, 00:28. Edited 3 times in total.

Reply 1 of 90, by brassicGamer

User metadata
Rank Oldbie
Rank
Oldbie

Well, all the numbers are there, so if anyone does have any significant criticism of the way you've presented your results, they can fix it their bloody selves. What would be good from you though would be some conclusions - who won? If anyone can't be arsed to look through the raw data, they should at least know what the headlines are. Good work - I'm looking forward to either corroborating or contradicting the results 😀

Check out my blog and YouTube channel for thoughts, articles, system profiles, and tips.

Reply 3 of 90, by Scali

User metadata
Rank l33t
Rank
l33t

Well, my conclusion is that although the K6-III+ is apparently the faster processor (eg, it scores the most CPU marks in 3DMark99 and scores best in Quake2 software rendering mode), it is let down by the rest of the system (chipset/motherboard), because the slower PII records the highest scores pretty much everywhere else.

http://scalibq.wordpress.com/just-keeping-it- … ro-programming/

Reply 4 of 90, by mrau

User metadata
Rank Oldbie
Rank
Oldbie

SR2 average<minimum?
in the serious sam file it says UT in the corner

now that was a ton of work; very useful, somewhat my experience with k6-2 as well, that it can slow down dramatically when the quality/reso is turned up
i wondered if an automated test culd be done with a very flexible engine for example quake2 tested in a window with resolution increments of 2 to see where this slowdown occurs

also: maybe place all results in 1 file, just more practical

Reply 5 of 90, by F2bnp

User metadata
Rank l33t
Rank
l33t

Thank you all for your kind words.

Scali wrote:

Well, my conclusion is that although the K6-III+ is apparently the faster processor (eg, it scores the most CPU marks in 3DMark99 and scores best in Quake2 software rendering mode), it is let down by the rest of the system (chipset/motherboard), because the slower PII records the highest scores pretty much everywhere else.

That would seem to be the case, however 3DMark uses 3DNow! to a great extent, where as on the Pentium II it's just using MMX. You can watch the Pentium III really pull away, thanks to SSE. That's why I think it is somewhat silly to use 3DMark as a comparison tool between different CPUs, it's best used as a metric to compare similar systems online, e.g. you've built a K6-III+ system and you want to make sure it is running properly so you check with someone else's score. Or you could just use to compare GPUs!

As regards Quake 2, without the 3DNow! patch, the K6-III+ leaves a lot to be desired. Once you install that patch though, it's just as fast as the PII 350, sometimes approaching the PII 400 even. Quake 2 was missing OpenGL performance for Crusher, I've added that in now!

I've always read that the FPU on the K6 CPUs is very weak compared to the P6 architecture. K6's FPU is actually slower per clock than Pentium MMX even. Combine that with the anemic memory performance and the K6-III+ is a disappointment in a lot of ways.

For what it's worth, if anyone's wishing to build the fastest K6-III+ system, go for the MVP3. It seems to be faster to my Aladdin V board and I'm sure that a couple of results that it turned in slower are just a matter of tweaking.

mrau wrote:

SR2 average<minimum?
in the serious sam file it says UT in the corner

Yeah, I kinda fucked up there, both of these are fixed now 😀.

Fun fact, I made all the charts using Office 97 on my K6-III+ while listening to MOD music in the background 😁. It seems that during the transfer they got a little fucked up, for example Quake 2 was missing a chart. I went through them once more and fixed all of these issues I believe. I've updated the .rar file!

Reply 8 of 90, by F2bnp

User metadata
Rank l33t
Rank
l33t
mrau wrote:

mods? can You share?

Share what exactly? I've got a collection tracker music such as videogame soundtracks (Unreal, Deus Ex, Jazz Jackrabbit 2, Agony etc) and various compositions from various musicians. You can find tons of it online, it used to be all the jazz in the 90's (before MP3 really caught on).

Tertz wrote:

Resume: K6-3+ 550 MHz ~ P2 400 MHz
acceptable for games up to 2000 year

Eh, I'd rather say K6-III+ 600 ~ PII 400
and even that would be pushing it somewhat as some games really tend to run like crap on the K6.

It was acceptable back then for up to 2000 as you're saying, just don't try to run Deus Ex on it 😁.

Reply 9 of 90, by Tertz

User metadata
Rank Oldbie
Rank
Oldbie
F2bnp wrote:

It was acceptable back then for up to 2000 as you're saying, just don't try to run Deus Ex on it

It's interesting to see fps comparision in 2000 year games. If there were such with internal benchmarks or exist external fps utility which will not ruin the performance on those CPUs. And used Geforce2-4, as most would use them as faster and cheaper than Voodoo5.

DOSBox CPU Benchmark
Yamaha YMF7x4 Guide

Reply 10 of 90, by F2bnp

User metadata
Rank l33t
Rank
l33t

I did include MDK2, Metal Gear Solid and Serious Sam (2001 release) for that reason. You're free to try more if you'd like, but I have to draw the line somewhere unfortunately. Realistically, I think most people would like a much more powerful system to enjoy such games with the high framerates (60fps +) that we're accustomed to these days.

GeForce cards should only really help performance in TnL enabled games, such as MDK2. 3Dfx cards are the better choice for building such a system for 1998-1999 games, as they have significantly lower CPU overhead in the drivers and work mostly without issues on these old AGP chipsets. GeForce cards will really shine with TnL games though, so it could be a fun project!

Reply 11 of 90, by Tertz

User metadata
Rank Oldbie
Rank
Oldbie
F2bnp wrote:

You're free to try more if you'd like

To build own mainly DOS machine I have no plans, as DOSBox does what I need still.
2 games of 2000 seem too few to show the situation, as the example of Deus Ex released in same 2000 shows the possibility of inconstancy of results. Earlier games work ok on Intel and AMD, while later games like Serious Sam are too hard for both. The border 2000 year for CPU comparision is the most interesting.

Realistically, I think most people would like a much more powerful system to enjoy such games with the high framerates (60fps +) that we're accustomed to these days.

30 fps are ok. The multimillion army of console users even don't suspect there can be more. 😀

GeForce cards should only really help performance in TnL enabled games, such as MDK2.

Taking into account CPU bottleneck - probably. But without real testing we can't be sure.

3Dfx cards are the better choice for building such a system for 1998-1999 games, as they have significantly lower CPU overhead

You used Voodoo5 which costs unreasonable money, so the results are practically less meaningful. While Voodoo3/4 may to give significantly lesser fps than Geforce 2-4, maybe even on K6.

DOSBox CPU Benchmark
Yamaha YMF7x4 Guide

Reply 12 of 90, by Scali

User metadata
Rank l33t
Rank
l33t
F2bnp wrote:

GeForce cards should only really help performance in TnL enabled games, such as MDK2. 3Dfx cards are the better choice for building such a system for 1998-1999 games, as they have significantly lower CPU overhead in the drivers and work mostly without issues on these old AGP chipsets. GeForce cards will really shine with TnL games though, so it could be a fun project!

Well, if you're comparing CPUs, the absolute graphics performance isn't that relevant. But the relative performance between the different CPUs is.
Perhaps the GeForce makes an interesting test case because it has more CPU/driver overhead. Question could be: does the K6-III+ handle this better than a PII/PIII or not?

http://scalibq.wordpress.com/just-keeping-it- … ro-programming/

Reply 13 of 90, by F2bnp

User metadata
Rank l33t
Rank
l33t
Tertz wrote:
f2bnp wrote:

GeForce cards should only really help performance in TnL enabled games, such as MDK2.

Taking into account CPU bottleneck - probably. But without real testing we can't be sure.

https://web.archive.org/web/20121114024004/ht … .com/show/707/7

Here are some MDK2 results using a Geforce2 GTS.

Tertz wrote:
f2bnp wrote:

3Dfx cards are the better choice for building such a system for 1998-1999 games, as they have significantly lower CPU overhead

You used Voodoo5 which costs unreasonable money, so the results are practically less meaningful. While Voodoo3/4 may to give significantly lesser fps than Geforce 2-4, maybe even on K6.

Yeah, Voodoo5 was an expensive card. I could have used Voodoo3 or Voodoo4, however these cards can be limiting with a few games such as Dethkarz, so I wanted to get rid of the GPU bottleneck entirely.

Voodoo3 usually produces the same results as the Voodoo5 and is more compatible. In fact, I'm using a Voodoo3 as my videocard on this system, the Voodoo5 feels much more at home on fast 440BX system (733Mhz and higher 😉). The Voodoo5 is just there to ensure that there never are any GPU bottlenecks, which is the same reason I wanted Imperious to test solely at 640x480, as some games can stress that card a bit more.

Reply 14 of 90, by Imperious

User metadata
Rank Oldbie
Rank
Oldbie

Thanks F2bnp for publishing the results, I certainly enjoyed contributing to this project.
I think the 2MB cache on my Epox MVP3G5 helped a bit with some results. Disabling it and relying
only on the 256KB in the K63+ definitely gave lower results.

I guess this project was about giving a general rather than definitive comparison between cpus. Yeah maybe
using a Voodoo5 against a Voodoo3 is a bit unfair, but it's not a apples to apples comparison anyway. I could have
tested my TI4200 but that wasn't the idea and anyway that video card is from 2002.

Atari 2600, TI994a, Vic20, c64, ZX Spectrum 128, Amstrad CPC464, Atari 65XE, Commodore Plus/4, Amiga 500
PC's from XT 8088, 486, Pentium MMX, K6, Athlon, P3, P4, 775, to current Ryzen 5600x.

Reply 15 of 90, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Very nice work, especially seeing so many game benchmarks used. Could you also include the Quake results without the 3DNow! patch? When I tested my system, I found the 3DNow! patch made only a small contribution.

It would be interesting to see how the K6-III-500+ at 5x83 on 430tx compares to the K6-III-500+ at 5x83 and 6x100 on MVP3, ALI, and SiS SS7 boards using identical hardware.

On average, would you say that the gaming performance of a K6-3+ 550 is similar to a PII-350, PII-400, or PII-450?

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 16 of 90, by F2bnp

User metadata
Rank l33t
Rank
l33t
feipoa wrote:

Very nice work, especially seeing so many game benchmarks used. Could you also include the Quake results without the 3DNow! patch? When I tested my system, I found the 3DNow! patch made only a small contribution.

It would be interesting to see how the K6-III-500+ at 5x83 on 430tx compares to the K6-III-500+ at 5x83 and 6x100 on MVP3, ALI, and SiS SS7 boards using identical hardware.

On average, would you say that the gaming performance of a K6-3+ 550 is similar to a PII-350, PII-400, or PII-450?

The Quake 2 results include non-3DNow! tests 😀.

I'd say it is mostly similar to a PII-350. On some games it can be a little faster and on some others it can be quite a bit slower (slower than a PII 333 even!). At 600MHz, the situation doesn't change too much, it does get a little closer to the PII 400 though.
I tried some 2D games like Desperados and Commandos and K6-III+ was faster than the PII 450 even. I couldn't really figure out how to benchmark these games properly though, so I left them out of the tests. Also, I tested Pyl, which is an old Polish FPS, although I once again could not benchmark it properly. I have played through the game on a K6-III+ 550 and it wasn't always a pleasant experience, with tons of variance in the framerate.
The Pentium II 333 was significantly better, minimum framerates improved dramatically.

K6-III+ is such an oddball 🤣 .

Reply 17 of 90, by melbar

User metadata
Rank Oldbie
Rank
Oldbie

Hhm, ok, when you say: K6-3+ 550 ~ PII-350

Back in the day, i had a Celeron466 (440BX system) with Geforce256DDR, overclocked stable @ 525MHz (7x75MHz). I assume, even with the little lower FSB, but higher clock than PII-350, less L2 cache but with full speed...
the Celeron should beat the K6-III+550 easily..?!

#1 K6-2/500 [VIA MVP3]
#2 Athlon1200 [VIA KT133]
#3 Celeron1000A [i815EP]
#4 P4-2800 [i845]

Reply 19 of 90, by feipoa

User metadata
Rank l33t++
Rank
l33t++
F2bnp wrote:
feipoa wrote:

Very nice work, especially seeing so many game benchmarks used. Could you also include the Quake results without the 3DNow! patch? When I tested my system, I found the 3DNow! patch made only a small contribution.

The Quake 2 results include non-3DNow! tests 😀.

OK, thank you. I must have been browsing too quickly because I didn't see them upon first glance.

Your results indicate:
23-25% increase in framerates in Quake II (software & MiniGL modes) when using 3DNow! optimisations.
4.6-6.6% increase in framerates in Quake III when using 3DNow! optimisations.

My results indicated only a 4-10% increase in Quake II frame rates, with the average being about 5.5% with non-3dfx cards. It was mentioned in another thread that you need to be using a 3dfx card to see these higher frame rates. When I tried to use a Banshee card with 3DNow! optimisations, 3dfx mode would not run for some reason.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486