I have a question about getting comparable Quake benchmarks:
Is it done by starting the game with "quake +timedemo demo2" which automatically hides the console, then shows the console when the results are in. About video, should one use this option: "-stdvid", which means 320x200. Should one set "video RAM cachable" in the BIOS? About audio: "-nocdaudio" or "-nosound"?

For my tests all I did was swap the CPU and do everything the same way with each test. I'm thinking I typed into the console manually for each timedemo but ran it twice to make sure the HDD didn't affect it. I don't think I had a sound card installed.

BIOS settings were the defaults as I recall.

Got it, thanks. 😀

Is it OK if I add a few benchmark results to this topic? Being a similar challenge at multiplier 4.0x, done at four different FSB speeds.

Edit 05-07-2011: Results for the same tests but with L2 cache disabled have been added, between brackets.
Edit 11-03-2017: Results for Tualeron have been added. The system was tested to give the already listed values for the coppermine core, but there was often a difference of about 0,6% in the score. Take this margin into account when comparing the Tualeron scores.

Pentium II, Deschutes SL2QF, half speed 512kB L2
200/_50 MHz: PCbench 31,0 // Quake _53,3 [19,1 // _45,3]
266/_66 MHz: PCbench 41,6 // Quake _71,1 [25,7 // _60,4]
400/100 MHz: PCbench 62,4 // Quake 105,6 [38,5 // _89,9]
533/133 MHz: PCbench 00,0 // Quake 000,0 crashes immediately...

Pentium III, Coppermine, QH31ES, full speed 256kB L2
200/_50 MHz: PCbench 29,5 // Quake _53,0 [20,6 // _47,2]
266/_66 MHz: PCbench 39,6 // Quake _70,5 [27,6 // _62,9]
400/100 MHz: PCbench 59,4 // Quake 104,9 [41,4 // _93,7]
533/133 MHz: PCbench 79,1 // Quake 138,5 [55,1 // 123,7]
Exact same results with SL3XG, the latter was used for the no-L2 test.

Pentium III, Tualeron, QID2QS, full speed 256kB L2
200/_50 MHz: PCbench 29,5 // Quake _54,5 [19,2 // _48,2]
266/_66 MHz: PCbench 39,3 // Quake _72,1 [25,5 // _63,9]
400/100 MHz: PCbench 58,8 // Quake 106,9 [38,3 // _94,9]
533/133 MHz: PCbench 78,5 // Quake 141,5 [51,1 // 125,7]

Pentium III, Tualatin-S, QEL8ES, full speed 512kB L2
200/_50 MHz: PCbench 32,4 // Quake _58,4 [19,1 // _47,4]
266/_66 MHz: PCbench 43,4 // Quake _77,8 [25,7 // _63,2]
400/100 MHz: PCbench 65,1 // Quake 115,6 [38,5 // _94,1]
533/133 MHz: PCbench 86,7 // Quake 152,6 [51,3 // 124,4]

Quake DOS v1.08 -stdvid -nocdaudio -nosound -nojoy +timedemo demo2
PC player bench DOS V1.04 640x400 8bpp LFB
System: GA-6BXC rev 2.0. with upgradeware BIOS; Voodoo3 AGP; 128MB PC133 SDRAM 3,3,3,6
Enabled: MTRR LFB WC.
NOT Enabled MTRR VGA WC, because it gives errors in a few games.
Although Quake would do 57,9 fps instead of 53,0 with the Coppermine at 200/50Mhz.

Surprise, the Coppermine is not faster per clock then the Pentium II with half speed cache!?

I wonder why it crashes at 533/133 on the PII and not the PIII. Is it because the cache is being overclocked past it's limits? Can the cache divider be adjusted like on the Slot A Athlons?

Yes, That Pentium II has cache modules for 333/2 MHz, and manages 400/2 MHz, but 533/2 is too much as expected. No cache divider options that I know of. I could retry with the L2 cache disabled, but it would not be fair.

Edit: when I opened up the processor card, which was not as easy as I hoped, it actually appeared to have 200MHz cache modules.

I remember 112x4.5 = 504 MHz being popular with some Deschutes chips. SL2W8 for example which I had. But going farther was too much for either cache or core or both, ya. 450 MHz was the most likely speed to work because that's what the cache chips were rated for.

It looks like Tualatin's full speed 512K cache is helpful as we all know. 😀

I would like to add a source of pure information about cpu internal behaviour and the corresponding speed effects:
http://www.azillionmonkeys.com/qed/cpuwar.html
Also check out the other pages from this guy, f.e. for the 7th gen CPUs.
I think this fits your benchmarks as background information very well.

Aw man, no Cyrix MII @ 133Mhz? 🙁

Tetrium wrote:

Aw man, no Cyrix MII @ 133Mhz? 🙁

Is it even possible to downclock them to that speed? The lowest multiplier I've seen on those is 2.5x which even if you could install it in a system with 60mhz bus equals 150mhz.

I'd settle for a 6x86 or 6x86MX at 133 MHz.

A bar chart might be beneficial for presenting the results of this post.

no sound option?

Hmm someone should run this on a modern Phenom II or i5 @ 133 FSB 🤣

sliderider wrote:

Tetrium wrote:

Aw man, no Cyrix MII @ 133Mhz? 🙁

Is it even possible to downclock them to that speed? The lowest multiplier I've seen on those is 2.5x which even if you could install it in a system with 60mhz bus equals 150mhz.

Now that you mention it, I didn't even consider that MII might not even have this multi (would need to underclock the fsb, influencing performance in a negative way).

Btw, there were S7 boards with 50Mhz FSB, but don't know from the top of my head when this FSB setting got removed 😊

You can move P3 slot1 cache to a P2 board if you like to run 533Mhz?

Or you can use a old program from Hoda´s wcpuid a L2 cache changer util for setting up its cache latency to higher then 2 or 3 or higher.
I used it once to set it to 1 to have higher benches, maybe it helps at reboot and modify jumper to fsb 133 and 533 mhz core speed succesfully? without the L2 cache crapping out at 266 by the higher CAS latency.
I think fsb jumper config can be made without powering down system.

Btw has anyone a Cyrix 5x86 133Mhz 486 machine at 50FSB with the software to enable the internal branch prediction like the pentium types have?

I haven't heard of any reports of a Cyrix 5x86-133 being able to load Windows at 3 x 50 MHz. As far as I know, all Cyrix 5x86-133's are of stepping 0 and do not work well with branch prediction in Windows, but Dos seems to work fine with a certain combination of register bits enabled (branch prediction included). Branch prediction works well, and in Windows, only on 5x86-100 units that have stepping 1.

For more information n which Cyrix 5x86 CPUs work well with which register bits, refer to the Cyrix 5x86 Register Enhancements Revealed (link below).

EDIT: Branch prediction seems to work on Stepping 0 chips as well, provided that RSTK and LOOP are disabled.

Pentium Classic (P54C) 133

3Dbench 1.0=90.9
MDKbench=59
PCPbench=14.6
Quake - timedemo demo1=28.1
timedemo demo2=29.1
timedemo demo3=26.7

Speedsys

Sandra 99

Thanks for the new results!

You should run those images through OptiPNG. 😁 Just a drag and drop ordeal.

@swaaye
from where i can download Sandra 2002?

swaaye wrote:

Thanks for the new results!
You should run those images through OptiPNG. 😁 Just a drag and drop ordeal.

Obviously the images are the highest quality JPEG now, but JPEG is unsuitable for images like this, which have many of single color areas. GIF and PNG are what one should use in that case, with GIF being limited to 256 different colors total.

So who feels like putting all results in a table? 😁