VOGONS


P2/P3 VS K6-3+ - A Great Battle Commences

Topic actions

First post, by F2bnp

User metadata
Rank l33t
Rank
l33t

For years the K6-2/3 CPUs, while beloved by a lot of people, have been called slow compared to what Intel was doing at the time, be it Pentium 2 and 3 or Celeron (not the Covington models).
However, without proper benchmarks we are left to merely guess how the fair against each other.
It is time to end the debate. It is time for change. 🤣

The hardware tested will be as follows:

On the AMD side:

K6-3+ 400, happily clocked at 550MHz (5.5x100) using 1.8V
Gigabyte GA-5AX (rev4.1)
256MB SDRAM (would have gone with 128MB, but I have no such sticks)
Voodoo 3 3000 AGP
AWE32
WD 10GB HDD

On the Intel side:

Pentium II 450 + Pentium III 500
Soyo 6BA+ III
256MB SDRAM
Voodoo 3 3000 PCI (don't have another AGP 😜)
SB16 or Live! or Vortex 2 (without any 3D sound obviously, although this could be a side-benchmark)
Something similar

A few points of interest:

-I've not yet begun the testing, I'm guessing I will start building in 2-3 days and then test for a few days. Feel free to suggest software and games to test!
-I will run the K6-3+ at 400 and 550 and the PII/III at 450, 500 and hopefully 550. I'm using the PII because I'd like to see what difference SSE makes (probably only on 3DMark stuff)
-Both boards feature ATA33, so no worries there.
-I'm thinking of running the Voodoo 3s at 3500 speeds (I know both are more than capable of achieving these clocks).
-I'd like to find out the performance hit with 3D audio, I should probably pick the Vortex 2 me thinks!!!

-Should I go with faster HDDs?

-Besides games, I will run 3DMark 99 and 2000.

Here are some of the games I'm thinking of benching with:

-Descent 3 -Revenant
-Ultima IX -Quake 1,2,3
-Half-Life -Unreal
-Shogo -Turok 2
-MDK 2 -Blade of Darkness
-Scorcher -AMOK
-Carmageddon -Mechwarrior 2
-Need For Speed 3 -Thief Gold

Please, feel free to comment on this, perhaps add games or software to run! I'd like to know what kind of tool I should use to measure FPS for those games that don't support time demos. An older version of FRAPS maybe?

IT IS FINALLY DONE!!!

Attachments

  • Filename
    Project P6 VS K6.pdf
    File size
    371.31 KiB
    Downloads
    883 downloads
    File license
    Fair use/fair dealing exception
Last edited by F2bnp on 2013-11-28, 14:26. Edited 2 times in total.

Reply 1 of 90, by Mau1wurf1977

User metadata
Rank l33t++
Rank
l33t++

Interesting project!

The thing is, when you have a Slot 1 platform, you aren't limited by 500 MHz. And the systems are very nice to work with. That's what makes this comparison more than just about performance.

The main benefit I see with Socket 7 is the ability to slow it down to a 386 and play games like Wing Commander.

My website with reviews, demos, drivers, tutorials and more...
My YouTube channel

Reply 2 of 90, by d1stortion

User metadata
Rank Oldbie
Rank
Oldbie

A shame that you don't have a PIII 450. If you run the PII at 500 it won't be apples-to-apples cuz of the higher FSB. I'm interested in seeing if SSE makes a practical difference and in this case with the two basically identical processors it would be a viable comparison.

For games I would test Unreal Tournament, which is a lot more demanding than Unreal in all regards. Would be nice to see how this game runs on a "budget" K6 rig from the same year. Why do you want to use 128 megs instead of 256?

What I would do is just do all the testing on one rig, note the results and then move parts such as video/sound cards to the other one just to ensure 100% comparability. Btw, I already did 3d sound comparisons and posted the results a long time ago, the average was just 2 FPS or so lower in UT (with a Live!).

Reply 4 of 90, by F2bnp

User metadata
Rank l33t
Rank
l33t
d1stortion wrote:

A shame that you don't have a PIII 450. If you run the PII at 500 it won't be apples-to-apples cuz of the higher FSB. I'm interested in seeing if SSE makes a practical difference and in this case with the two basically identical processors it would be a viable comparison.

It is indeed a shame, but not a terribly big loss, I doubt 50MHz and FSB difference will show a big advantage. We'll see!

d1stortion wrote:

For games I would test Unreal Tournament, which is a lot more demanding than Unreal in all regards. Would be nice to see how this game runs on a "budget" K6 rig from the same year. Why do you want to use 128 megs instead of 256?

It is eh? Me thinks UT will kill the K6! I just think 256MB are an overkill, these machines are just fine with 128.

d1stortion wrote:

Btw, I already did 3d sound comparisons and posted the results a long time ago, the average was just 2 FPS or so lower in UT (with a Live!).

I'm more interested in seeing how the K6-3+ is faring in this compared to a PIII which shouldn't have too much of a problem.

Reply 6 of 90, by idspispopd

User metadata
Rank Oldbie
Rank
Oldbie

I would either use ISA sound cards in both machines or PCI sound cards in both machines. IIRC ISA sound cards slow down games more than PCI sound cards. Or you could turn off sound if the game supports it.
3dNow! like Davros suggested is a good idea. The Quake II 3dNow! patch comes to mind.
FRAPS sounds good, that's what Putas used for his benchmarks.

Which OS are planning to use? 98SE? Or XP? I'm not sure which OS would run the games faster, might even depend on the CPU and the game.

Reply 8 of 90, by d1stortion

User metadata
Rank Oldbie
Rank
Oldbie
idspispopd wrote:

I would either use ISA sound cards in both machines or PCI sound cards in both machines. IIRC ISA sound cards slow down games more than PCI sound cards.

1 FPS difference is not noticeable. Maybe with an even weaker CPU, but UT would be not very playable in the first place then.

Reply 9 of 90, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

The K6 chips were only competitive with the PII in the early days...around 1998 or so. They weren't faster though, they were just good value for the money. Once Intel started ramping up the clock speeds they left the K6 in the dust. I remember when K6-III came out I was pretty excited, except that they got pretty poor yields and were never able to get the clock speeds high enough to compete with the p3. If I recall correctly the K6+ chips came out after tha Athlon was announced...I think around the time the P3 hit 1GHz. I think the K6+ chips were always intended to be used as upgrades and in budget systems. If they had come out a year or two earlier that would have really been something...except that the Super7 platform royally sucked.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 10 of 90, by idspispopd

User metadata
Rank Oldbie
Rank
Oldbie
d1stortion wrote:
idspispopd wrote:

I would either use ISA sound cards in both machines or PCI sound cards in both machines. IIRC ISA sound cards slow down games more than PCI sound cards.

1 FPS difference is not noticeable. Maybe with an even weaker CPU, but UT would be not very playable in the first place then.

OK, maybe the difference is not that big. I was just suggesting to reduce the number of variables when comparing both systems.

Anonymous Coward wrote:

The K6 chips were only competitive with the PII in the early days...around 1998 or so. They weren't faster though, they were just good value for the money. Once Intel started ramping up the clock speeds they left the K6 in the dust. I remember when K6-III came out I was pretty excited, except that they got pretty poor yields and were never able to get the clock speeds high enough to compete with the p3. If I recall correctly the K6+ chips came out after tha Athlon was announced...I think around the time the P3 hit 1GHz. I think the K6+ chips were always intended to be used as upgrades and in budget systems. If they had come out a year or two earlier that would have really been something...except that the Super7 platform royally sucked.

PII/PIII has better memory performance at the same bus speed which matters more with higher multipliers. (I think there were some interesting results in the 133MHz challenge for PPro and PII with a multiplier of only 2x and the PII cache running at FSB clock speed.)
The K6+ chips were intended for notebooks. People tried to get them as upgrades but they were hard to get, and the don't run on all SS7 mainboards.

Reply 11 of 90, by F2bnp

User metadata
Rank l33t
Rank
l33t

It is my aim to run 3Dnow! optimised games. Quake 2/3, Descent 3, MDK2 are optimised. I'll also add the following games:

- Drakan Order of the Flame
- Hostile Waters
- Evolva
- Outcast

Unreal Tournament is also a good contender, perhaps I'll remove Unreal for its favor. Also, I'll be using ISA cards initially and a Vortex 2 or Live! later, to test 3D sound. Initial testing will not include 3D sound performance. OS is obviously 98SE! 😁
I'd like to try out more SVGA games like Mechwarrior 2 (which I will be running at 1024x768). Any suggestions?

Anonymous Coward, all of this is very true. The purpose of this whole thing, is to truly see how much slower a K6-3+ (and K6-2/3 on the same clock, the difference is very small) is compared to a similarly clocked Pentium 2/3. I'm pretty sure there are games where the K6 will be just as fast or almost as fast as a Pentium. The Aladdin V (GA-5AX's chipset) is actually quite good, I've never had any experience with the Via MVP3. Obviously, the 440BX is awesome 😀

Reply 12 of 90, by elianda

User metadata
Rank l33t
Rank
l33t

I would prefer to have at least one K6-2CXT benched, as the K6-2 was much more mainstream than the K6-3+ (some 300,350 or 400 MHz K6-2). Also some slower P2 should be included like the 266 MHz one as it was kind of cheaper entry level to the P2 platform, mostly on a 440LX board with 66 MHz FSB. This way you have several performance points for comparison. As already written a 450 MHz P3 as the fastest CPU would be sufficient.
You have to be aware that early K6-2s have no MTRR support and K6-3+ have already extended 3DNow! support, so the K6-3+ is already a rather special CPU in the K6 line.
Be sure to have the K6 systems correctly setup, regarding MTRR, cacheable area, the right driver versions, game patches.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 13 of 90, by swaaye

User metadata
Rank l33t++
Rank
l33t++

I'd like to see a Covington Celeron (or a simulated one via disabled L2). These CPUs existed to compete with K6 chips and I recall them being recommended over K6 for gaming, for good reason.

Reply 14 of 90, by F2bnp

User metadata
Rank l33t
Rank
l33t

Unfortunately, I don't have a K6-2 anymore. I do have a Pentium 2 233 though, which I should be able to easily run at 266 and I think it could hit 300.

I had no idea the K6-2/3+ had extended 3Dnow! support, interesting.
Elianda, could you please remind me what MTRR was exactly and how I can set it up correctly, as well as the cacheable area? What are the optimal settings?

Since I have a Klamath 233, I can do Covington. In fact, shouldn't it run at 300 rather easily? Isn't L2 the limiting factor in overclocking usually?

Reply 15 of 90, by mwdmeyer

User metadata
Rank Oldbie
Rank
Oldbie

I have a Covington Celeron, a bunch K6 2s and a K6-3 400MHz. Pretty sure I also have a Pentium II 400. I'll see if I get some time to do some benchmarks 😀

Vogons Wiki - http://vogonswiki.com

Reply 16 of 90, by swaaye

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

I had no idea the K6-2/3+ had extended 3Dnow! support, interesting.

It's only partial. They have the 5 new 3DNow DSP instructions but lack the integer SSE additions.

Reply 17 of 90, by d1stortion

User metadata
Rank Oldbie
Rank
Oldbie

Another area of interest might be emulators. A PII 350 can already be enough to play ePSXe decently, while a PIII 500 might do fullspeed in many games. I would say SSE makes a big difference in such applications, and consequently K6 will be crushed mainly due to its weak floating point performance.

Reply 19 of 90, by GeorgeMan

User metadata
Rank Oldbie
Rank
Oldbie
d1stortion wrote:

Another area of interest might be emulators. A PII 350 can already be enough to play ePSXe decently, while a PIII 500 might do fullspeed in many games. I would say SSE makes a big difference in such applications, and consequently K6 will be crushed mainly due to its weak floating point performance.

FYI:
Just set up a mini PC for my father to play 3D PSX games via ePSXe (e.g. Crash Bandicoot 3).
Found a microATX QDI BrillianX V (intel 440BX) mobo w/ Vibra sound chipset: http://www.motherboard.cz/mb/qdi/BrillianX%20V.htm
Combined it with Riva TNT2 M64 32MB AGP and 192MB RAM.

Came with PII 333 (FSB66). Not playable at all with DX6 plugin, major fps drops.
Switched to PIII 450 and downgraded to epsxe 1.7 from 1.9.
Playable full speed at "fast" settings, but really bad colors!

Then I lastly tried to overclock it to 500MHz (112x4.5) and downgraded to ePSXe 1.6.0.
It is playable at "Nice" settings, no lag at all! 😁
But I guess it's about ~95% CPU usage :p

But who cares, pretty happy now as I have 2 slot1 machines running on purpose @ 2013! 😁 😁

1. Athlon XP 3200+ | ASUS A7V600 | Radeon 9500 @ Pro | SB Audigy 2 ZS | 80GB IDE, 500GB SSD IDE2Sata, 2x1TB HDDs | Win 98SE, XP, Vista
2. Pentium MMX 266| Qdi Titanium IIIB | Hercules graphics & Amber monitor | 1 + 10GB HDDs | DOS 6.22, Win 3.1, 95C