Bussing around with a G3 Beige XPC106 Memory controller upgrade

[PowerPC74]

I dropped in a nice little socket that makes swapping crystals a breeze.

View attachment 2505658

I received a 17.734MHz crystal, which is the next step up from 16MHz, as 17MHz crystals don't seem to be available.

The absolute best bus speed I have been able to attain is 94.7MHz as reported by Gauge Pro, with just one stick of RAM. Adding in more sticks brings the bus speed down as previously (to 89-91MHz or thereabouts).

Changing the reference crystal should scale things up linearly, so I would expect a bus speed of 103MHz. As before, I'm hitting some kind of ceiling.

Still, it's not far off the 100MHz mark I'm aiming for and I've seen no instability or weirdness despite PCI running at almost 38MHz.

View attachment 2505660



I'm pretty sure it's the first one – excessive PLL load causing the clocks to sag. Gauge Pro does seem to settle down as you previously observed on another machine. But removing RAM affects the bus speed a lot more conspicuously.

I wonder whether bumping the clock generator's voltage slightly might help.

Are all the ram sticks PC100/PC133?

Are you using mixed pairs, or all the same speed and timings?

I doubt the XPC100 will do 133MHz, someone said the tried with the B&W and it was highly unstable, but you may get to 100MHz if Mac-io and the PCI bus is stable.

Also, I wonder how Gage Pro is able to calculate the bus speed?

 

[pete1]

Are all the ram sticks PC100/PC133?

Are you using mixed pairs, or all the same speed and timings?

Here’s what I’m using. A (almost) matched set of PC133 RAM. One of them is a CL2, the other two are CL3. At PC100 speeds, which I’m not even reaching, they’ll all run at CL2.

Also, I wonder how Gage Pro is able to calculate the bus speed?

I also wonder this. It does seem pretty dependable, but I’d be interested to know how accurate it is.

 

[PowerPC74]

Here’s what I’m using. A (almost) matched set of PC133 RAM. One of them is a CL2, the other two are CL3. At PC100 speeds, which I’m not even reaching, they’ll all run at CL2.

View attachment 2506194



I also wonder this. It does seem pretty dependable, but I’d be interested to know how accurate it is.

Could be a ram timing issue on an unexpected bus speed, or we can hope. What it the PLL clock generator multiplier to PCI Bus speed, exactly?

I asumme it's somewhere close to 2.14x the FFL?

You may want to try the 30Mhz PCI setting and the 35Mhz, to see if the ram issues persist.

 

[pete1]

Could be a ram timing issue on an unexpected bus speed, or we can hope.

My take on it is that the clock generator isn’t strong enough to drive multiple outputs much above design speed. The CPU and RAM clocks likely need to be buffered, but this would be extremely difficult to implement. And if you’re going to do that, you might as well design a new clock generator.*

There might be some ways to increase headroom slightly. I’m looking into this.

What it the PLL clock generator multiplier to PCI Bus speed, exactly?

I asumme it's somewhere close to 2.14x the FFL?

33.3333/14.31818=2.328

You may want to try the 30Mhz PCI setting and the 35Mhz, to see if the ram issues persist.

It doesn’t work any better on those, unfortunately.

My eventual aim is still to get that 3x multiplier working for 33.33MHz PCI and 100MHz CPU bus. As I currently understand it, setting the Grackle to 3:1 mode is only half of the solution. The SC608 clock generator also actually needs to output a 3:1 CPU : PCI clock, which it will never be able to do. So, a replacement clock generator* is needed for that, something that we can program ourselves. I believe there are a number of off-the-shelf solutions, which require looking into.

 

[PowerPC74]

My take on it is that the clock generator isn’t strong enough to drive multiple outputs much above design speed. The CPU and RAM clocks likely need to be buffered, but this would be extremely difficult to implement. And if you’re going to do that, you might as well design a new clock generator.*

There might be some ways to increase headroom slightly. I’m looking into this.



33.3333/14.31818=2.328



It doesn’t work any better on those, unfortunately.

My eventual aim is still to get that 3x multiplier working for 33.33MHz PCI and 100MHz CPU bus. As I currently understand it, setting the Grackle to 3:1 mode is only half of the solution. The SC608 clock generator also actually needs to output a 3:1 CPU : PCI clock, which it will never be able to do. So, a replacement clock generator* is needed for that, something that we can program ourselves. I believe there are a number of off-the-shelf solutions, which require looking into.

So you think we maybe able to replace the SC608 clock generator and get 3x out of one?

 

[pete1]

So you think we maybe able to replace the SC608 clock generator and get 3x out of one?

The tray loader iMac uses the SC608 to generate PCI at 33.33MHz - the other outputs are tied to ground. An MPC932 drives the RAM and CPU bus clocks, taking the PCI clock as an input and multiplying it by 2x.

So I think we should do something like that - keep the SC608 but just use it for PCI. Then send its PCI output to a to a clock driver which supports 3x. The MPC932 can do 3x and runs up to 120MHz - but it doesn’t have enough outputs for 3 sticks of RAM. It would be fine for experimentation, and a solid starting point since it is tried and tested.