VOGONS


First post, by Stesch

User metadata
Rank Newbie
Rank
Newbie

Hello everyone,

while this is my first post, i have been visiting vogons for the last 5 to 6 years very frequently. My first 486 was a present from my boss with a 486SX inside working at a staggering 33Mhz.
When I started my first game (i only played dos games on a pentium 3 PC before), i was a little unhappy to see that it was way too slow to be enjoyable. So I started upgrading to a dx2, a dx4, and an AMD 5x86 later on.

After i had my AMD 5x86 running at 160Mhz (I thought I had reached the end of my journey, how naive), I've found feiopa's guide about the Cyrix 5x86 CPU registers and his reports about running an IBM 5x86c at 2x 66Mhz/133 Mhz. Seeing his impressive benchmarks, I new it was time for me to replace that AMD with a Cyrix and go on to a hole new adventure loosing half of my sanity (I blame you exclusively for this feiopa!)

Getting 120Mhz was easy enough, since my first Cyrix rated at 100Mhz would cope quite well at 120, even with the default 3.6 volts. But it always intrigued me to try to go for more 😀
Problem is, i had no luck getting the 120Mhz version of the 5x86 to run at 150Mhz, my IBM 5x86c would boot but it was never even remotely stable and I didn't (and still) don't want to overvolt more than i am confortable with.
I also tried 2x 66Mhz on two of my SiS 471 boards (my favorite chipset), but was never happy since something was always off, like the cache being way to slow with the needed wait states, nearly all VLB graphic cards having graphical glitches, and on one mainboard, the keyboard controller just passed away temporary (at least I asume it was the reason since the intro of screamer was still running fine but i could not skip it / exit). Yes, i *could* use an PCI 486 mobo with an 1/2 divider, but for me, I wanted to stay with VLB. VLB on a 486 mb just feels somehow 'good' for me.

So I started thinking about how to reach for something in between the boring 120Mhz and the too fast 150Mhz, and found an old website that talks about replacing the crystal oscilator for overclocking. (But it was not for 486 PCs, but some weird japanese PCs I know exactly nothing about). At first, I was skeptical because I thought that it would cause a lot of complications, since for reaching lets say 133Mhz, i would need an FSB of ~44Mhz instead of 40Mhz and it would screw up the sound, floppy drive and perhaps more. But then, nearly every sound card uses an own crystal anyway (or 3 crystals, like my SB16 card), exactly like the IDE controllers I have, so I gave it a try.

IMG-4370.jpg

Desoldering the old crystal and replacing it with a socket strip (not sure if this is the correct term, in german it's called 'buchsenleiste') was no problem. Just break the pins you dont need away, it also does not matter if it is too long like mine since there is enough space on the mainboard anyway. I ordered two additional crystals with the stock 14,31818Mhz rating in case i would damage them while unsoldering (My soldering skills are, let's say, expandable)

IMG-4375.jpg

these are the crystals i had ordered to toy around with:

12,2880-HC18 -----slower speeds for 50Mhz selection
12,7500-HC18
13,0000-HC49U-S
13,8750-HC49U-S
14,3180-HC18 ----- default speed
14,7456-HC18 -----higher speeds for 40Mhz selection
15,0000-HC18
16,0000-HC18

I ordered some that are slower than the stock 14,31818 ones, so i could try to jumper 50Mhz FSB and then reach less with the modified crystal. In the end, i found the 12,75Mhz variation (~3x44.52Mhz, ~134Mhz) to fit the best.

Some benchmarks and the BIOS settings below, but keep in mind that when it shows 150Mhz this is incorrect! I think speedsys and most other programs rely on the now messed up timer (too slow, thus giving more time for the benchmarks) and it may very well be that the RAM and cache throughput are also displayed incorrectly.

IMG-4379.jpg
IMG-4378.jpg
IMG-4380.jpg
IMG-4383.jpg
IMG-4390.jpg

Disadvantages:

Replacing the crystal worked like a charme for me, but not all sound cards worked after the replacement. My ESS Audiodrive had a distorted 'click' after playing samples, but replacing it with a SB16 value solved the problem. (I intented to use the SB16 anyway, the Audiodrive just was within easy reach on my test cardbox 😀 ). Also, I had problems with my VLB IDE controller, when running the 'dir' command under dos, sometimes it would hang up the system. After replacing the card with an ISA controller, it worked just fine. One downside was that my 1MB of l2 cache that worked flawless @40MHz would not cope with the tightest wait states, so I got my box with my ISSI IS61C1024N cache modules (yes, they are all from china and most of them do not fulfil their speed rating in the slightest) and lost my last bits of sanity replacing these buggers for hours... But in the end, I was able to get it working stable with the tightest wait states at ~44.5Mhz!

In order to do it properly, you would need to reroute the clock signal for the keyboard controller as far as I understand since else the PIT chip will run at the wrong speed (to slow in my case) resulting in wrong timers and a wrong system clock but to be honest, i found the impact to be not too big since many games (from my feeling especially later ones) do not use the pit chip for timing anyways. (But I don't know the details here, so please don't be angry if I talk rubbish)

Advantages:

When using a 45Mhz FSB, your cache and ram will also get faster of course (at least if you don't need additional waitstates to get it running) which gives a nice performance boost. The Cyrix 5x86 with most CPU registers enabled scales impressively with every additional mhz and I am very happy with the results! (Screamer finally runs completely fluid on "high" settings 😀 ) I tried to run the quake demo with playdemo demo1 but it gives me no score, perhaps this is not the correct version (it says 1.08)? If anyone wants to see some benchmarks, feel free to request.

Please be patient in case I made writing errors, english is not my native language but i give my best.

Best regards, stesch

Reply 1 of 2, by rmay635703

User metadata
Rank Oldbie
Rank
Oldbie

The old Micronics ISA motherboards you could install any oscillator you want even a variable one.

I did follow the lead of someone else to modify a more modern motherboards base clock but the date/time and floppy controller were all jacked up.

There may be other ways of injecting an independent clock into the cpu that doesn’t match the motherboard FSB.

I would take a look at the old variable speed upgrade solutions for XT/AT systems, would need an interposer and a means to inject your own clock.

I do not know how well this solution works on anything past a 286 but it was an interesting solution for 8088/80286/80287 chips to clock them above the motherboard FSB without messing up other motherboard functions.

Reply 2 of 2, by Stesch

User metadata
Rank Newbie
Rank
Newbie
rmay635703 wrote on 2025-06-09, 21:46:
The old Micronics ISA motherboards you could install any oscillator you want even a variable one. […]
Show full quote

The old Micronics ISA motherboards you could install any oscillator you want even a variable one.

I did follow the lead of someone else to modify a more modern motherboards base clock but the date/time and floppy controller were all jacked up.

There may be other ways of injecting an independent clock into the cpu that doesn’t match the motherboard FSB.

I would take a look at the old variable speed upgrade solutions for XT/AT systems, would need an interposer and a means to inject your own clock.

I do not know how well this solution works on anything past a 286 but it was an interesting solution for 8088/80286/80287 chips to clock them above the motherboard FSB without messing up other motherboard functions.

From what i know, even some old 486 mainboards still had a discrete crystal for the CPU clock generation (but i don't have any of that kind) so there it should be easy to replace it without messing up the time etc.. But of course the performance of them is not great and you would have a hard time getting newer CPUs running like the Cyrix 5x86.

How 'modern' was that mainboard where you tried the replacement? I assume it already had an onboard IDE and Floppy controller? IIRC, feipoa mentioned that he could increase the crystal speed by about 0.7 Mhz before the floppy would stop working. In my case, it's no problem since the ISA IDE controller where the floppy is connected to has its own 14,31818 Mhz Crystal 😀

To get the time working properly, i think you would have to grab the 14,31818 Mhz clock signal from somewhere else and hotwire it to clk pin of the keyboard controller while disconnecting the original trace , but to be honest, im not sure if that is the correct way to do this. Perhaps if someone else with more experience than me could confirm this, i would give it a try 😀