Brute forcing stuff is not always the best way though.

If VESA is faster, then use VESA modes.

If we are going to limit it to hardware only, then graphics benchmarks need to be off the table or we limit to only a specific video card and a very specific set of settings, drivers, etc. if you want to artificially limit it like that.

Also, no hardware mods whatsoever either... cause it was never meant to be used like that.

As i mentioned early:
Ideally the test stays constant.
Goal is to augment the environment it runs inside, to achieve better performance, stability, or whatever else is on your mind.
Any sw/hw mods are just fine.
No ?

The Serpent Rider wrote on 2021-05-17, 14:59:

I use PhilsComputerLab dosbench, like Peter 'cpu-galaxy' for testing, so it is comparable

We're squeezing as much FPS as possible at this point and using VESA is still a legit way to obtain higher score.

Are all those parameters temporary or could it be set permanent, so cpu-galaxy is not aware that he set it before and it is still active?

That's the point i am trying to make.
If we don't measure with the same ... well ... measure, then the whole retro performance chasing hobby becomes even more pointless. 😀
So we start to doubt every achievement - was it based on "improved" tests, or not ?
Anyway.

Are all those parameters temporary or could it be set permanent

Depends. Phil's pack bat files will drop some variables after restart, including any changes to vid modes. Normal exe run or more streamlined bat file will keep variables permanent. Maybe except "vid_wait", which will change after switching video card.

As long as people post the exact settings they are using, I see no problem. That way others can replicate the settings.

That should always be done anyway, along with the version number of the programs used for benching as well as the resolution and driver versions.

Thanks for all the great info! especially from feipoa.

I've just started building and testing a setup consisting of the Biostar MB and a AMD x5 133mhz@180mhz, and could use some help getting the best performance out of it. The objective is to OC/optimize to the max stable config and then leave it there.

The setup boots W98SE@180mhz with 256kb double-bank 15ns cache and EDO@60ns or FPM@70ns. I've played around with timings, fsb etc. but none of it seems 100% stable in Win. No stresstesting is done yet.

Please advice, but what I've planned on doing next is; check and adjust CPU-voltage, check and adjust FSB, mount and test peltier (I have 2-3 of them), buy 1x 64mb, 60ns FPM RAM, buy 10ns cache (512kb double bank).

Any tips for the setup? Does anyone have the mentioned parts for sale or know where to get? I'm currently following ebay.

Thanks!

CPU is air cooled? What is exact model and date of it?

Where are You going to buy 10 ns cache?
Ia have bought 10 ns module from china and they works like original 15 ns 😀

You can try switch from WB to WT cache in BIOS settings.

60ns FPM RAM

If motherboard can work with EDO - use EDO modules. In general they'll overclock better.

Any luck with a 486?

@lukas12p, yes CPU is cooled by a big cooler and powerful server-fan, it doesn't get warm at all on idling, and just lukewarm when booting/using Win. This cooling setup is temporary, so a better permanent one will come into play.
Model is 8433-UUD or whatever it's called, rev. 3 and I think date was march 95(?)
The 10ns cache some people say is better, some say it's not, who's got testes to back up the claims? 😀

I tried some more last night with feipoa's manual, tested out 4v cpu-voltage, 50x4 and 66x3 fsb but machine didn't start. Also took out half the cache since it was not the kind in feipoas' manual, and left in he other half which was the right kind (taking it from 256 to 128kb + tag), although I don't know if it makes any difference in this OC experiment. The ram I teste with is NEC 60ns EDO, 16MB.

Where to go from here? I'm going to test with peltier, and try and get som 60ns FPM RAM and maybe some 10/12ns cache, any advice?

I don't have 8433-UUD motherboard, so can't give You specific advises.
I use LS-486 and EDO ram are faster than FPM. 60 ns is ok
L2 10 ns cache might help to get lower timings, but there is many fake ones that works like 15 ns or slower.
If You can get peltier to cool CPU to 0 deg. C it would be great. I have some Chinese ones that doesn't work well..
At temperature about 10 C I can boot to DOS @3x66 MHz, at about 6 C run Quake, but can't run Doom.

Yah, Doom is most sensitive to instability, Windows98 next, and Quake quite insensitive.... by the time you get non-heatsoak related crashes in Quake, it's just as likely to freeze halfway through a dir command in DOS.

I rearranged some things in my 5x86 case today; printed out a 3.5" CF-IDE bay (by HanJammer I believe) so I rerouted a few cables. I ended up with extreme system instability. Soft reboot lockups, Windows not loading, Quake crashing etc. It took me about an hour to figure out that one of the IDE cables was stopping the CPU fan. Overheating CPU was the culprit. I was lucky to not damage my IBM 5x86-100. Everything back to normal now, running great at 120MHz with an MX86200 PCI 😀

Can't help making myself a mental "to do" list reading this thread, though I probably wouldn't get to doing until later in the year. So far it goes something like.. dig out the motherboards, make one run at 75mhz, corral all the DX4s, find out if any are 16kB WB, assemble high pressure blower system, make siliconimage PCI SATA RAID card work on 486 mobo... test all PCI Rage cards for dos pixel banging... and it gets crazier from there.

Seems that CPU-Galaxy has sparked a flame here 😀

Original instigator is feipoa though.

The Serpent Rider wrote on 2021-05-25, 11:47:

Original instigator is feipoa though.

Yeah, he deserves all credit - his work on CPU registers in particular is amazing.

H3nrik V! wrote on 2021-05-25, 10:27:

Seems that CPU-Galaxy has sparked a flame here :)

We already had data for an Am5x86 at 200 MHz 10 years ago; just there is renewed interest. Seems like "pics or it didn't happen" has been recently replaced with "vids or it didn't happen".

True that. No doubt that feipoa's work is brilliant! But it seems that CPU-Galaxy has reborn it .. 😀 I'm for one, intrigued to dig out the 486 machine and some x5's, peltiers and maybe even some water 😁