14.818MHz is still used to the day for frequency generator ICs, ISA, and video cards, and lot of cards has this.

16MHz I doubt it, could be turbo mode off. Chipset set in CMOS setting determines the division frequency like 8MHz, 10MHz and 12MHz.

32KHz crystal is for clock and date time keeping.

I want to overclock Isa bus. Which oscillator would you replace - the 14 or the 16? I assume 16 as I can select clk/2 or clk/3 in bios. Thus coming to the most common 8mhz. Anyway i can be wrong.

Any hints? Pic attached.

Damn. Or does the bios setting indeed direct to the cpu oscillator since 25mhz /3 = 8,33. maybe not since clk/2 is standard setting which would mean hard overclock out of the box. Very uncommon.
Otherwise it could be the option for the sx16 shipment of the board.
You know I don’t want to solder sth without a need. Any hint how to find out what to do?

What is going on then?, Didn't tell us the history of this motherboard and what's the issue?

Cheers,

I'm confused, too. Why soldering, at all?

I thought the ISA bus clock is usually set in CMOS Setup..

With that big 4 pin quartz oscillator (50 MHz) being relevant for chipset/CPU clocking (actual used frequency is halve of this, 25 MHz).

That's why it's socketed, often, held by a zip tie sometimes. It's being replaced if the CPU is changed.

An ISA bus frequency of 10 or 12 Mhz is not too awkward, I think.
So CPU+Mainboard Clock (25 MHz) divided by 2 (12,5 MHz) or 3 (8,33MHz) makes sense to me.

Some fast 286 PCs from the 80s ran at 12 Mhz directly still, including the 'chipset' and ISA bus

That's because the 286' Front Side Bus (FSB) used to be the bus that connected each component with each other.
The ISA bus or AT bus was nothing different than the 286 signal lines on an edge connector.

Since this is a 386SX motherboard (a 286 derivative),
it maybe can handle that higher clock speed itself just fine ?

Just like it's 286 cousin ? 286 chipsets were still "recycled" for use with 386SX CPUs, after all..
Sometimes directly, sometimes with a little modification.
That's why 386SX boards were so affordable, afaik.
They had kept intelligent features, too, like shadow memory or Expanded Memory support.

Edit: To be precise, the 286 and 386 have an CLK2 input (just called CLK on 286).
This input will divide the incoming operation frequency by two (as coming from, say, a 50 MHz TTL oscillator).
Because it's merely measuring a rising/falling edge signal, I suppose.
It's up to the mainboard if it's doing the same (it does normally).
So it may run at 25 MHz or 50 MHz here, depends on the motherboard.

Edit: I rewrote the text partially. Hope it's better readable now. Though I still have my doubts. 😅

Edit: But what if that SX motherboard and its BIOS were originally made with the 386SX16 CPU in mind ?

The 386SX16 @16MHz was a popular flavor of the 386SX, after all.

Then clk/2 would be perfectly fine as a default setting for the BIOS (= 8,33 MHz). 🙂👍

Thanks guys at first. Let me be a bit more precise:

I want to overclock my ISA bus on the board for benchmark purpose. I just want to know where to apply the change for that.

I was thinking the 50Mhz oscillator is for the cpu frequency only whereby the ISA bus frequency is set independently via the 14 or 16MHz oscillator. This is what you can read very often in the internet (see my initial post).

From mathematical point of view you can explain each root oscillator being responsible for the Isa bus speed 😀

Looking forward to your further help.
Thanks

just experiment a little, swat 16 for another 14 and see if vgaspeed went down

I will go for that!

Don't change the 14MHz oscillator. Some devices in your PC depend on it for timing, and you might end up with a non-functioning floppy drive for example. If your ISA clock is based on that crystal, then there's nothing you can do, unless your BIOS gives you the option to toggle between that or clock dividers.
The 16MHz crystal may be for your ISA clock, it may be for de-turbo mode. Maybe both. Trace the pins and see where it goes.

Thx again. I will tackle the 16 only. Turbo was down to 12 mhz I think thus cpu oscillator / 4 instead of /2. I assume… 😀

Keep u posted

Since some of you mentioned benchmarks..
I just did remember something funny.

"Oh, and you might have heard of the "Chang Modification" from late 1988 when a
Taiwan-based company called Tienwei Computers declared they had produced a 40 MHz 80286 motherboard.
The media saw it running benchmarks, and it all looked pretty legitimate - the world was astounded.
Soon after they discovered it was all smoke and mirrors - he "gamed" the computer clock in such a way that
the Landmark Speed Test gauge calculated performance incorrectly, making a system look about four times more powerful than it really was.
This was achieved by making the DOS timer run slower, so effectively what is measured as "10 seconds" by a benchmark program would in reality take 20 seconds if you halve the DOS timer frequency. "

Source: https://www.dosdays.co.uk/topics/get_a_286_ru … e_a_386_pt1.php

I hadn't heard this story before. That's pretty funny.

Sooo, first feedback:

Replacing the 16Mhz oscillator didn’t change anything. I checked with vidspeed * and thruput. Basic stuff to show memory and video bus performance / bandwidth.

So I will go for the 14 now carefully

Marco wrote on 2022-10-07, 18:16:

Sooo, first feedback:

Replacing the 16Mhz oscillator didn’t change anything. I checked with vidspeed * and thruput. Basic stuff to show memory and video bus performance / bandwidth.

So I will go for the 14 now carefully

When you modify the 14.318 crystal, all benchmarks will misreport the speeds. You will need to compare against your wristwatch / phone timer instead to find out what really happens. The 14.318 crystal provides the "reference" the speed of your is compared against. So putting a 16MHz crystal where the 14.318 belongs will decrease most benchmark scores, because the same operations executed at the same speed take more 16MHz cycles than they took 14.318MHz cycles.

Understood. So I better skip this one. Strange so it really seems that Isa speed is connected to the cpu speed only. No dedicated Isa osci. Pity. I hoped to be able to tune the old one even more

Marco wrote on 2022-10-07, 18:41:

Understood. So I better skip this one. Strange so it really seems that Isa speed is connected to the cpu speed only. No dedicated Isa osci. Pity. I hoped to be able to tune the old one even more

There are two ways to couple the host processor to the slower ISA bus. Either both are clocked separately, and everytime the host processor needs to access the ISA bus, the request needs to be put in a buffer waiting for the next ISA clock to pick it up, and when the request is finished, the answer is put into a response buffer, waiting to be picked up at the next host clock. Putting the data into buffers and waiting for the next clock, and getting this done safely even if clock edges nearly conincide is difficult and you might need to add extra timing margin for data integrity. This kind of design is called an "asynchronous" design.

On the other hand, if the ISA clock is directly derived from the host clock, the host chipset knows exactly when the next ISA clock edge will come, and can operate the ISA bus in lockstep with every n'th clock of the host clock. This is called a synchronous design, and is considerably easier and cheaper to design. That's why many chipsets just support this kind of synchronous design. If the target ISA clock is an integer divider of the host clock (like 25MHz/3 = 8.33MHz), a synchronous design is likely both the easiest and the fastest. On the other hand, at 20MHz host clock, with a divider of 3 you would run the ISA bus at just 6.7MHz, which is slow, and with a divider of 2, you would run it at 10MHz, which is out-of-spec. An asynchronous design that allows 8 to 8.33 MHz ISA clock is likely faster than a synchronous design at 6.7MHz.

If your board runs at a synchronous design using 25MHz/3 for the ISA bus, there might be a configuration option in the chipset (possibly via jumper, solder bridge or software initialization) to run at half the processor clock (intended for 16MHz processors). If you find a way to toggle that option, you will get 12.5MHz ISA at 25MHz processor clock. If the chipset is also meant to support 33MHz processors, a /4 option is also likely present. ISA@12MHz will work with a lot of cards, but be careful with some sound cards and old IDE drives.

Great and very interesting. Will check for a soldier bridge as there are no jumpers available for that. But tomorrow when there is sun again 😀
Thx

What VLSI chipset does your board use? There are data sheets for the VLSI 82C320/VLSI 82C331 combination (TOPCAT). The 82C320 is actually software-programmable. You have the option to choose between synchronous operation or asynchronous operation. In either case, the ISA slot frequency is the oscillator frequency divided by 2, 4, 6 or 8. If your board uses this chipset, its the BIOS that chooses the configuration. Possibly you can adjust the divider using the BIOS setup utility.

Hi,

It’s indeed the vlsi 311 386sx mainboard using the VL82C311 chip.

Picture: http://www.amoretro.de/wp-content/uploads/201 … otherboard2.jpg

The only thing I can adjust in advanced setting is DMA clock with SCLK/2 or /3. but this doesn’t have any affect so far on benchmark performance.

Ideas on how else to program - even this chipset? Thanks again