- Glide fits good to this time.
- Make sure you configured MTRRs right, either by BIOS, FASTVID and Setk6v3.
- Put a bit of work into fast working AGP settings on the Ali chipset. (3DMark2001SE could be run to check, it usually crashes soon if something is not working)
- I would stay at FSB 100 to have this identical for the Pentiums and K6-III.
- Make sure you have the right patches for Quake (e.g. 3dnow! with K6-III)
- Split the system and setup description and condense the results in a table not larger than one screen. This helps comparisons greatly.
- AWE32 and SB16 will basically be identical with the Windows Games.

elianda wrote:

- Glide fits good to this time. - Make sure you configured MTRRs right, either by BIOS, FASTVID and Setk6v3. - Put a bit of work […]
Show full quote

- Glide fits good to this time.
- Make sure you configured MTRRs right, either by BIOS, FASTVID and Setk6v3.
- Put a bit of work into fast working AGP settings on the Ali chipset. (3DMark2001SE could be run to check, it usually crashes soon if something is not working)
- I would stay at FSB 100 to have this identical for the Pentiums and K6-III.
- Make sure you have the right patches for Quake (e.g. 3dnow! with K6-III)
- Split the system and setup description and condense the results in a table not larger than one screen. This helps comparisons greatly.
- AWE32 and SB16 will basically be identical with the Windows Games.

-Already taken that into consideration.
-Taking a note of that! Good call!
-Already done.
-I don't really agree with this one. Pentium II 350 would have been awesome to test, Pentium II 333 is a nice addition, I think it helps us pinpoint the K6's performance more easily.
-Done!
-Ideally, I would love to have charts, the same way Phil does, but I have never done any of these, so I'll probably seek help after I've gathered all the data (right now I'm making notes by hand, as I've done before in the past).
-That's the idea on those SB cards, do you reckon I should maybe aim for PCI cards? I've got a few Live! I could use, but I doubt it's worth it.

Needless to say, thanks for the input!

I have an unlocked Deschutes P2-333. If you'd pay shipping, I'll give it to you. I live in the US. It is the cpu I tested with in my project. My motherboard only support 66Mhz FSB, so I can't confirm it will run at 350, but I know it ran from 333 all the way down to 133. BTW, when I bought it, it did not come with the stock HSF. I jury-rigged one from a socket 7 heatsink, but just want you to be aware of it.

edit: just noticed you could buy a P2-350 for $9 shipped on ebay, so maybe that's a better option for you. But my offer stands. 😀

Even if you don't do the charts, I'd still use a spreadsheet to take notes with and record results. Much less likely to have human recording errors and such. I have an old KVM so my DOS and WinXP machines can be switched back and forth with a double-click of the ScrollLock key. This helped speed up recording of benchmark results for me.

I also agree that staying with stock FSB (doing whatever possible, if at all possible, with the multiplier) is the only way to produce correct results and not burn such a great benchmarking project.

(Really? A PII 350 is that close to a K6-3? No numbers handy, but I'd expect the K6 to be a lot faster and not just slightly faster in some occasions. That's news for me 😀 )

clueless1 wrote:

I have an unlocked Deschutes P2-333. If you'd pay shipping, I'll give it to you. I live in the US. It is the cpu I tested with in my project. My motherboard only support 66Mhz FSB, so I can't confirm it will run at 350, but I know it ran from 333 all the way down to 133. BTW, when I bought it, it did not come with the stock HSF. I jury-rigged one from a socket 7 heatsink, but just want you to be aware of it.

edit: just noticed you could buy a P2-350 for $9 shipped on ebay, so maybe that's a better option for you. But my offer stands. 😀

Even if you don't do the charts, I'd still use a spreadsheet to take notes with and record results. Much less likely to have human recording errors and such. I have an old KVM so my DOS and WinXP machines can be switched back and forth with a double-click of the ScrollLock key. This helped speed up recording of benchmark results for me.

Thanks for the offer buddy, but I'm in Europe, so I'll probably try finding one locally. Plus, I wouldn't want to grab such a nice specimen from you, an unlocked Deschutes is something worthy of keeping! 🤣 Honestly though, thank you for the offer!
My 333 is a later revision and as such is multi locked 🙁.

konc wrote:

I also agree that staying with stock FSB (doing whatever possible, if at all possible, with the multiplier) is the only way to produce correct results and not burn such a great benchmarking project.

(Really? A PII 350 is that close to a K6-3? No numbers handy, but I'd expect the K6 to be a lot faster and not just slightly faster in some occasions. That's news for me 😀 )

I'll see what I can do about that, and yes, last time I tested, a K6-III+ 550 was trailing the Pentium II 400 by a little bit, at least as far as games that were unbiased were concerned. Quake 2 is faster with the 3DNow! patch, but then you've got MDK2 and Descent 3, I believe, where the 400 really veers off.

This might be extra work, but it would be awesome for the K6 to have unpatched and patched results. With patched I mean 3DNow! patches or any other tweaks that help it out.

Same goes for the Voodoo, they are known to do well on the K6. Including a Nvidia card, like a TNT2 or GeForce256 would be awesome.

Because this is a lot of work, maybe start off with the most interesting CPUs to see if a picture emerges, then work from there.

F2bnp wrote:

Thanks for the offer buddy, but I'm in Europe, so I'll probably try finding one locally. Plus, I wouldn't want to grab such a nice specimen from you, an unlocked Deschutes is something worthy of keeping! 🤣 Honestly though, thank you for the offer!
My 333 is a later revision and as such is multi locked 🙁.

No problem. 😀 I bought it specifically to test its performance with L1 cache disabled, then when it arrived I was surprised it did not have the HSF on it. The seller's photos conveniently did not show that side, and there was no text saying "HSF not included" or somesuch. I mentioned this to the seller (without asking for a refund) and they just refunded me without further communication. I would have taken a partial refund happily, but we didn't even get into a discussion about it.

Make sure vsync is disabled. Test low resolutions so you are not GPU bottlenecked.

Maybe see if you can find a benchmark that will give you statistics on frame rate lows. Serious Sam SE has a nice benchmark tool but that game may be too demanding. Unreal's timedemo function gives average, RMS, instant, lowest and highest. UT does something somewhat similar IIRC.

ALI AGP Utility might be useful for dealing with ALI AGP settings. It was quite useful with P5A and Voodoo3 for making DVD playback possible.

Use a game called breakneck aka Nice 2 it has a build-in benchmark and supports voodoo.

I have a K63+ 550MHZ on a Epox EP-MVP3G5 motherboard, 600mhz is possible for some benchmarking but not stable.

I also have a Athlon XP system that can be clocked down to 450mhz so could throw that into the ring with some benchmarks, although the 512k cache
will likely bump up results a bit too much. It is running on a pc133 sdram board that technically only supported up to 1.4 Athlon Thunderboard.

I can't help with Voodoo5 benchmarking but do have a ti4200, mx440, fx5500, tnt2 pro,voodoo3.

PhilsComputerLab wrote:

This might be extra work, but it would be awesome for the K6 to have unpatched and patched results. With patched I mean 3DNow! patches or any other tweaks that help it out.

Same goes for the Voodoo, they are known to do well on the K6. Including a Nvidia card, like a TNT2 or GeForce256 would be awesome.

Because this is a lot of work, maybe start off with the most interesting CPUs to see if a picture emerges, then work from there.

You're right on the money here, although Quake 2 is the only game I can think of where there was a 3DNow! patch, unfortunately. 3DNow! optimization was moved to the driver side for the most part, which is why we didn't get any other patches.

I'd like to try out an Nvidia card, preferably a GeForce2 or something like that. However, besides the few games that can take advantage of the HW T&L like Max Payne, MDK2 and maybe Quake 3, I can't quite think of any other game that would benefit in the slightest. The K6 is so severely limited, that 3Dfx will run faster because of Glide (obviously) and lower driver overhead overall. Still, it'd be kinda cool trying to see if Max Payne becomes playable with a GeForce card.

swaaye wrote:

Make sure vsync is disabled. Test low resolutions so you are not GPU bottlenecked.

Maybe see if you can find a benchmark that will give you statistics on frame rate lows. Serious Sam SE has a nice benchmark tool but that game may be too demanding. Unreal's timedemo function gives average, RMS, instant, lowest and highest. UT does something somewhat similar IIRC.

ALI AGP Utility might be useful for dealing with ALI AGP settings. It was quite useful with P5A and Voodoo3 for making DVD playback possible.

Good points. Vsync is disengaged at all times, don't want that double buffered V-Sync destroying my results 😵 . I will be testing mainly at 640x480 to show the full effect of CPU bottlenecking, plus 1024x768 where possible, on the off-chance that I catch the Voodoo5 holding the system back and monitor the effects this has on overall performance. A benchmark that shows minimum framerates would be godsend in such a case.

Again, perfect point bringing up frame rate lows. I like Unreal and UT, however the only bench I know that is spot on is UTBench and even that is a taking to the extreme sometimes. I don't have much love for flybys as they aren't really representative of actual gameplay. Nevertheless, I'll probably record results from both Unreal and UT flybys, plus UTBench. Do you know of anything else that has a similar setup to UTBench?

I've fooled around with the ALi AGP utility, but haven't managed to get much out of it. I think it probably was very useful in that DVD playback experience you had, but it's usually tweaking with the BIOS directly that will get you results in games (like that one option I had to open up a thread about, just a few months ago 🤣 ) . Performance for my little K6-III+ is tuned to the max honestly, since I've spent countless hours getting it to run as fast as possible 😊 .

meljor wrote:

Use a game called breakneck aka Nice 2 it has a build-in benchmark and supports voodoo.

Cool! Will totally check it out, thanks!

Imperious wrote:

I have a K63+ 550MHZ on a Epox EP-MVP3G5 motherboard, 600mhz is possible for some benchmarking but not stable. […]
Show full quote

I have a K63+ 550MHZ on a Epox EP-MVP3G5 motherboard, 600mhz is possible for some benchmarking but not stable.

I also have a Athlon XP system that can be clocked down to 450mhz so could throw that into the ring with some benchmarks, although the 512k cache
will likely bump up results a bit too much. It is running on a pc133 sdram board that technically only supported up to 1.4 Athlon Thunderboard.

I can't help with Voodoo5 benchmarking but do have a ti4200, mx440, fx5500, tnt2 pro,voodoo3.

That's mighty cool of you, we can directly compare MVP3 and Aladdin V in that case! Perhaps you can try 5.5x105 for that 577MHz setup I mentioned? Should probably be alright, I can't get mine to even boot at 600MHz, but it is 100% stable at 577. Go figure.
Athlon is nice actually, but it will totally destroy any opponents with that full-speed 512k cache as you mentioned 🤣 . It's probably faster than the first Coppermines at that clock.

Most of the games I mentioned produce the exact same performance on the Voodoo3, as they do on the Voodoo5, because of CPU bottlenecks. So, you could totally use that Voodoo3. PM me for details if you're interested more, I'll give you lengthy instructions on how I benched every game! 🤣

A lot of really cool replies and ideas, thanks so much for the interest you're showing, makes me happy that people care about this 😀.

Underclock the K6-III? I had a K6-III that was rated as a K6-III 333mhz on a 66mhz bus. I'd test them at the same speed's as the Pentium II from 333 onwards. Also would be interesting if you had some Celeron's thrown in as well. As the Celeron was the K6's compeitor not the Pentium

F2bnp wrote:

That's mighty cool of you, we can directly compare MVP3 and Aladdin V in that case! Perhaps you can try 5.5x105 for that 577MHz […]
Show full quote

Imperious wrote:

I have a K63+ 550MHZ on a Epox EP-MVP3G5 motherboard, 600mhz is possible for some benchmarking but not stable. […]
Show full quote

I have a K63+ 550MHZ on a Epox EP-MVP3G5 motherboard, 600mhz is possible for some benchmarking but not stable.

I also have a Athlon XP system that can be clocked down to 450mhz so could throw that into the ring with some benchmarks, although the 512k cache
will likely bump up results a bit too much. It is running on a pc133 sdram board that technically only supported up to 1.4 Athlon Thunderboard.

I can't help with Voodoo5 benchmarking but do have a ti4200, mx440, fx5500, tnt2 pro,voodoo3.

That's mighty cool of you, we can directly compare MVP3 and Aladdin V in that case! Perhaps you can try 5.5x105 for that 577MHz setup I mentioned? Should probably be alright, I can't get mine to even boot at 600MHz, but it is 100% stable at 577. Go figure.
Athlon is nice actually, but it will totally destroy any opponents with that full-speed 512k cache as you mentioned 🤣 . It's probably faster than the first Coppermines at that clock.

Most of the games I mentioned produce the exact same performance on the Voodoo3, as they do on the Voodoo5, because of CPU bottlenecks. So, you could totally use that Voodoo3. PM me for details if you're interested more, I'll give you lengthy instructions on how I benched every game! 🤣

A lot of really cool replies and ideas, thanks so much for the interest you're showing, makes me happy that people care about this 😀.

I Do have some Durons but don't really want to mess with the heatsink again as I have to pull the motherboard out to change it. I cannot do 105x with this motherboard as it only supports 112, 124, 133
but caches have to be disabled for those as the 2MB motherboard Cache will not run higher than 100mhz. Last time I tried messing with that I completely screwed up Win98se and had to reinstall.

candle_86 wrote:

Underclock the K6-III? I had a K6-III that was rated as a K6-III 333mhz on a 66mhz bus. I'd test them at the same speed's as the Pentium II from 333 onwards. Also would be interesting if you had some Celeron's thrown in as well. As the Celeron was the K6's compeitor not the Pentium

At these speeds it's just going to be immensely slow, not really willing to try it, even madness has its limits 🤣 . As far as the Celerons go, that's a really neat idea, unfortunately I don't have any slot 1 Celerons or even slockets. I could turn the L2 cache off to simulate a Covington though.

Imperium, that's cool, you can give us your results for K6-III+ at 550MHz! I'll talk to you soon about it.

What is the point of testing both K6-III+ 550 and K6-III+ 577? Little to no difference, drop the overclocked one.
Your game selection is shooter heavy, I would like to see more diversity. Quakes are beaten to death, if you want to do them spice it up with optimized 3dnow libraries.
Might be interesting to test video driver with optional 3dnow optimalization (I know at least Ati and Nvidia had them around 99-2001). Maybe even SSE.
Are you going to benchmark with sound? I would prefer that over usual settings, but maybe use same sound card in all rigs, just to be sure.

Neat project! I've been wondering about this lately as well.

I've been working on my "ultimate" socket 7 system for some time now (non-super socket 7, mind you). It's a K6-3+ overclocked to 500MHz on 83MHz system bus (Asus P55T2P4), paired with a Voodoo 3 2000 overclocked to 175MHz. It's severely CPU bottlenecked and I've been wondering if it's working right. I remembered my K6-2 system back in the day being not much slower.

Built a K6-2 400 system with Voodoo 2 to compare it to, and yep! The K6-2 really is so much slower, it's pathetic 🤣

I really like socket 7 as a platform and all AMD processors for it, but one has to manage their expectations.

Really looking forward to your results.

Do you have any decent PCI video cards lying around? It would good to see if PCI limits performance at all on these systems, the fastest cpus anway.

Putas wrote:

What is the point of testing both K6-III+ 550 and K6-III+ 577? Little to no difference, drop the overclocked one. Your game sele […]
Show full quote

What is the point of testing both K6-III+ 550 and K6-III+ 577? Little to no difference, drop the overclocked one.
Your game selection is shooter heavy, I would like to see more diversity. Quakes are beaten to death, if you want to do them spice it up with optimized 3dnow libraries.
Might be interesting to test video driver with optional 3dnow optimalization (I know at least Ati and Nvidia had them around 99-2001). Maybe even SSE.
Are you going to benchmark with sound? I would prefer that over usual settings, but maybe use same sound card in all rigs, just to be sure.

Hey Putas! The point of testing 550 and 577 is arguably the same as doing 400 and 450 or 450 and 500 on Intel. I'd prefer to do 550 and 600, but since I can't make the system stable at 600 (6x100), I figured I might as well do 577 (5.5x105) which should be very close.

As far as games go, could you recommend any games you'd like to see? I know you've tested dozens and dozens of games on your website, but how did you go about testing titles with no benchmark tools? Could you point me towards specific games with specific testing methods?

As far as 3DNow! libraries and drivers go, from what I gather the only few games with 3Dnow! DLLs are Quake 3 and maybe a handful OpenGL titles, most of them based on idtech 3. I could include Heavy Metal FAKK2, but I think that would absolutely kill both Intel and AMD systems, so I didn't see the point of doing it. Then again, I have Serious Sam in the test list...
As far as drivers go, I believe 3Dfx drivers started including 3DNow! optimizations from a certain point and onwards, so using the latest drivers should have the most up to date optimizations (assuming they ever updated them after first introducing them anyway).

Yes, I will be testing with sound, I always hate it when people disable it xD. I mean what's the point of trying to show how each system performs in real situations and then disabling the sound?

oerk wrote:

Neat project! I've been wondering about this lately as well. […]
Show full quote

Neat project! I've been wondering about this lately as well.

I've been working on my "ultimate" socket 7 system for some time now (non-super socket 7, mind you). It's a K6-3+ overclocked to 500MHz on 83MHz system bus (Asus P55T2P4), paired with a Voodoo 3 2000 overclocked to 175MHz. It's severely CPU bottlenecked and I've been wondering if it's working right. I remembered my K6-2 system back in the day being not much slower.

Built a K6-2 400 system with Voodoo 2 to compare it to, and yep! The K6-2 really is so much slower, it's pathetic 🤣

I really like socket 7 as a platform and all AMD processors for it, but one has to manage their expectations.

Really looking forward to your results.

Thanks for the nice words man! Yes, K6-2 400 is just too slow hahaha 😵 .

dirkmirk wrote:

Do you have any decent PCI video cards lying around? It would good to see if PCI limits performance at all on these systems, the fastest cpus anway.

I actually do have a Voodoo3 3000 PCI and of course a couple of Voodoo2 SLI cards, however I know from previous testing, on the Super 7 platform, that PCI will not make a difference, as far as the V3 goes anyway. By the way, I've seen Voodoo5 AGP vs PCI differences only on high clocked Athlon XPs, you're usually hitting other limits/bottlenecks way before bus speed ever plays any significant role.

Hallo members 😁 Think this is my first post 😀
I am so anxious to see serious sam results on K6!
It is so slow on my K6!!
GTA 3 runs super awesome! Quake 3 bench in the 70s fps
3Dmark 2000 I get just over 4700 that makes me feel serious sam must be playable but its around 15-20fp 🙁
I have not used any of the programs like k6speed yet as I do not understand how to use it 🙁