Powermate SX/20vi: Oscillator replacement and interesting side-effects thereof...

[super-sama]

I have a FedEx Powership 2, pretty much an NEC PowerMate SX/20vi special-ordered by FedEx to distribute as their Powership system for small to large scale shipping room solutions.
It comes in a different case with a different PSU, front panel connections, and ISA riser card (normally has 4, FedEx machine has 2).
Comes stock with a i386SX/20, I've added in a 387SX 16-25MHz, and 12MB of RAM for a total of 14MB (2MB onboard, parity.)

After a botched attempt at upgrading it with a TI486SXLC2 (-G50, 3.3v part, oops) ...and dropping in an am386SX-40 for cheaper than a 20 or 25 would have been, I'm at the point of prepping the machine for a clip-on Cx486SRx2-25/50MP, which I should have gotten in the first place. the chipset on the board is a WD76C10, initially meant for 286 systems. it can be found in all variants of the Amstrad Mega PC and Sega Teradrive, ranging from 286 to a TI486SLC. the max bus speed it can officially support is 25MHz, so I can't crazy-up the bus speed and enjoy just the am386sx on its own, lest I damage the chipset from an unnecessary overclock. the solution is the Cyrix clip-on, and the utilities to turn on the 1K of cache.

I removed and socketed the main bus oscillator, and upgraded it from 40MHz to 50MHz for a bump in speed from 20 to 25MHz, since I've read someplace that the Cyrix upgrades are a bit picky on the machine actually being the proper bus speed for the selected chip. There was one oscillator to the left I initially thought was for the onboard WD90C11A Paradise graphics chip, and it was sitting there at 48MHz. so being me and being stupid at 3AM, I socketed that one too and dropped in another 50MHz so they were matched.

Aside from a nice bump in throughput on all counts with only one 50MHz oscillator installed, the second one made it a bit weird, I guess?
video performance didn't change in the amount I thought it was going to, instead what I got was a very slight bump in access times, RAM bandwidth, and... floppy drive speed??
when the second one is installed, the floppy drive seems to seek faster, as in the head makes a slightly faster sound when it seeks. not sure if the motor is turning faster too.

since finding this out, I've replaced the second back down to 48MHz, until I figure out just what's going on with that, since I don't want to cause some sort of compatibility issue with the drive running faster than it should be even with a difference of 2MHz. the system wasn't locking up in either case and with the second 50MHz oscillator installed, I did a stress-test for a few hours and ran memtest without any hiccups.

Here's the Speedsys results:

Stock:

25MHz (50MHz), second 48MHz crystal stock:

25MHz (50MHz), second crystal at 50MHz:

and finally, a pic of the board in question:


The far-left cut off oscillator is 64MHz, and I've since confirmed it's probably for the WD90C11A chip instead, after digging through the datasheet for it.


so if the 40MHz (now 50) is the bus speed oscillator, what the heck is the 48MHz one controlling the speed of?
if it helps anything, the cards I have in it at the moment are a 3C509B Etherlink III with an XT-IDE bootrom, and an SB16 CT1740.
floppy drive is a Samsung SFD-321B, and I can take pics of the motherboard with both oscillators removed if needed, since I have a second parts board I've removed them on already.

 

[GiGaBiTe]

Motherboards back in those days usually cascaded clocks off of a main clock source, usually the CPU clock. If the floppy disk is seeking faster, it could be the result of either the ISA bus which it is connected to is running faster; Or it could be since the CPU is faster that the floppy drive is able to push data faster. Unless you have some weird floppy drive that does CLV rather than the normal CAV, the motor isn't going to spin faster or slower.

Disk controllers exhibit the same phenomenon, faster CPU and bus clocks meant they could push more data. I have a 486 with a turbo button that has an IDE drive in it. When the CPU is in the faster clock mode, you can hear the drive activity get faster than when in the slower mode.

There was never really a fixed standard for ISA bus clocks, most manufacturers usually ran it at the 4.77 MHz if it was an 808x or 7-10 MHz on later systems using a clock divider. Though there were a few odd ducks which ran the ISA bus beyond that, up to 16 MHz, which usually caused compatibility issues with cards as that was too fast.

 

[super-sama]

Motherboards back in those days usually cascaded clocks off of a main clock source, usually the CPU clock. If the floppy disk is seeking faster, it could be the result of either the ISA bus which it is connected to is running faster; Or it could be since the CPU is faster that the floppy drive is able to push data faster. Unless you have some weird floppy drive that does CLV rather than the normal CAV, the motor isn't going to spin faster or slower.

Disk controllers exhibit the same phenomenon, faster CPU and bus clocks meant they could push more data. I have a 486 with a turbo button that has an IDE drive in it. When the CPU is in the faster clock mode, you can hear the drive activity get faster than when in the slower mode.

There was never really a fixed standard for ISA bus clocks, most manufacturers usually ran it at the 4.77 MHz if it was an 808x or 7-10 MHz on later systems using a clock divider. Though there were a few odd ducks which ran the ISA bus beyond that, up to 16 MHz, which usually caused compatibility issues with cards as that was too fast.

yep so doing more digging, per this tidbit here, for the WD76C30 super IO chip onboard...
"Integrated clock generation circuitry uses the 48 MHz input signal to generate the 1.8462, 3.072, and 8.0 MHz clocks used internally for the two serial ports, a 9.6 MHz Signal used for the keyboard controller and floppy controller, a programmable duty/frequency clock for the 80287 coprocessor, and a 16 MHz clock for driving the WD76C10 Single Chip AT Controller, and floppy controller."


so that explains that. the 48MHz crystal drives the 76C30, which in turn provides the clocks for the 76C10 and 76C20. however the 76C10 still relies on the 50MHz CPU clock oscillator aside.

 

[GiGaBiTe]

So in this case it looks like the floppy controller clock is asynchronous from the CPU, so the faster floppy seeks is likely because the CPU is able to move data around faster.

 

[super-sama]

yep, exactly. I've since ziptied both the 50MHz and 48MHz oscillators in place and it will stay there.