New Awesome and Easy Way to Bump a Wombat to 40MHz

[Phipli]

I've found a new way to overclock Wombats.

If you have a 33MHz Q650 or a Q800...

Step 1 - bridge J29 with solder, it is next to the CPU.

That's it.

Your Wombat will now boot at 40MHz.



The machine photographed above has a 16.6MHz clock fitted at G3!

More detail.

So it turns out that the chip U71 is only there to select between two clock sources, all shipped Wombats, it is hard wired to route the clock output from G3, the 12.5 or 16.6MHz clock, to U21, which doubles it's frequency and sends it to the CPU... But it turns out that if you fit J29, it selects an alternative 20MHz clock from elsewhere on the logic board! If you fit jumper pins, you can switch between 40 and 33MHz by adding and removing the jumper.

If you want to improve stability, remove resistor R152, or ideally...

The ultimate setup (for machines that started at 25 or 33Mzhz) :

Fit J29
Fit J28
Fit a 330 ohm resistor to R233
Fit a 1.2k ohm resistor to R152
Make sure R151 is not fitted.

J28 selects between 33MHz hardware timings (no jumper) and 40MHz hardware timings (with jumper).

At this point, you can power down, change the jumpers (you can reach them without having to remove the mezzanine) and power back up at a different processor speed!

 

[Phipli]

@joshc @cheesestraws @ everyone with a wombat.

 

[Phipli]

The following were done at 40MHz with 33MHz timings (the result is that the video is overclocked, and I'm more likely to have stability issues, but I was just testing the basic "just add J29 and nothing else" scenario. With R152 removed, video scores would drop back down to normal for a Q650.









The hard disk is just a 230MB Quantum, so I didn't really need to include it.

 

[Phipli]

View attachment 1000016996.mp4

 

[eharmon]

Wow, that's an incredible find. So 33 and 40MHz Wombats really are using the exact same board with identical components besides different jumper/resistor configurations and memory speeds. That's really curious!

I wonder if overclocking the 20MHz clock source is more stable, as well. That said, I bet they're tapping the same source as NuBus? Might be a bad idea to push that .

 

[zigzagjoe]

Wow, that's an incredible find. So 33 and 40MHz Wombats really are using the exact same board with only different jumper/resistor configurations and memory speeds. That's really curious!

I wonder if overclocking the 20MHz clock source is more stable, as well. That said, I bet they're tapping the same source as NuBus? Might be a bad idea to push that .

Nice find for sure, @Phipli.

While I don't have a schematic, I'd assume the 20mhz clock is a divided signal from that osc near the front of the board?

 

[Phipli]

Nice find for sure, @Phipli.

While I don't have a schematic, I'd assume the 20mhz clock is a divided signal from that osc near the front of the board?

That's what I assume, I didn't follow it back to source.

 

[Phipli]

Wow, that's an incredible find. So 33 and 40MHz Wombats really are using the exact same board with identical components besides different jumper/resistor configurations and memory speeds. That's really curious!

There are a few configurations - only 25 and 33MHz machines shipped, and they came with 80, 70 and 60ns RAM fitted. I think I remember seeing that the Q800s have slightly more aggressive RAM timings, even compared to the same speed Q650.

This jumper will even set a 25MHz to have a 40MHz CPU clock, but the timings would still be set for 25MHz which would probably crash it, so you'd want to... Remove R151 in that situation or preferably both R151 and R152.

I wonder if overclocking the 20MHz clock source is more stable, as well. That said, I bet they're tapping the same source as NuBus? Might be a bad idea to push that .

I anticipate it would be worse. It is used for other subsystems, while G3, the one we've traditionally replaced is for the CPU specifically. This is just an easy way to get to 40MHz, but doesn't help going beyond.

The way to make the most overclockable Wombat is to set the timings to 40MHz (the fastest it has), remove onboard RAM, use a 50 or 60ns RAM SIMM, and possibly replace the U21 chip with an MC88916DW80, although I'm not certain that is required.

 

[Phipli]

That's what I assume, I didn't follow it back to source.

@zigzagjoe
I had a look, the 20MHz clock comes out of the Nubus controller, which the 40MHz clock is attached to, so it is basically a divide by two of the Nubus clock.

It also feeds into the SCSI and Ethernet.

 

[eharmon]

I think I remember seeing that the Q800s have slightly more aggressive RAM timings, even compared to the same speed Q650.

They do, yeah. And fitted with 60ns stock instead of 70 to match. It appears the theoretical stock 40MHz Wombat would have also used 60ns but with looser timings, based on the data in ROM.

I anticipate it would be worse. It is used for other subsystems, while G3, the one we've traditionally replaced is for the CPU specifically. This is just an easy way to get to 40MHz, but doesn't help going beyond.

Yeah I figure it's a real bad idea. Fun to think about though.

I had a look, the 20MHz clock comes out of the Nubus controller, which the 40MHz clock is attached to, so it is basically a divide by two of the Nubus clock.

That's clever. And confirms just how bad an idea overclocking it probably is.


Once you jump J29, does the 20MHz source get routed to serial/ethernet as well? They definitely seem to run fine at 40, so I am guessing it still affects the entire bus?

 

[Phipli]

Once you jump J29, does the 20MHz source get routed to serial/ethernet as well? They definitely seem to run fine at 40, so I am guessing it still affects the entire bus?

20MHz is always routed to ethernet and SCSI. Serial doesn't directly use the 20MHz, I'm not sure where the serial clocks are derived without more searching.

J29 just adds an extra path from the 20MHz to the clock Doubler, it doesn't change anything that is already on that clock.

 

[eharmon]

20MHz is always routed to ethernet and SCSI. Serial doesn't directly use the 20MHz, I'm not sure where the serial clocks are derived without more searching.

J29 just adds an extra path from the 20MHz to the clock Doubler, it doesn't change anything that is already on that clock.

Huh. So problems with ethernet after overclocking are more likely due to insufficient waits on the faster processor bus and not the ethernet controller itself running too fast? I always thought it was the latter, but the former makes sense.

 

[Phipli]

Huh. So problems with ethernet after overclocking are more likely due to insufficient waits on the faster processor bus and not the ethernet controller itself running too fast? I always thought it was the latter, but the former makes sense.

They're are clock related reasons on the 610, but on the 650, yeah, sounds like it will always be due to software timings.

Did I read apple had to release an update because they broke ethernet timings on stock wombats?

 

[eharmon]

They're are clock related reasons on the 610, but on the 650, yeah, sounds like it will always be due to software timings.

Did I read apple had to release an update because they broke ethernet timings on stock wombats?

There goes my theory on why my 800 seems to have broken onboard ethernet at stock clocks all of a sudden...sigh.

I'm pretty sure they did fiddle with timings over the years, but I'm not sure they ever broke compatibility with stock machines. Certain drivers run worse for me overclocked than others. Roughly I've noticed (and hopefully wasn't imagining things) that later drivers seem to do worse than earlier ones at deadlocking. I suspect they tightened the timings as they were squeezing faster performance out of the later drivers.

 

[Phipli]

There goes my theory on why my 800 seems to have broken onboard ethernet at stock clocks all of a sudden...sigh.

I'm pretty sure they did fiddle with timings over the years, but I'm not sure they ever broke compatibility with stock machines. Certain drivers run worse for me overclocked than others. Roughly I've noticed (and hopefully wasn't imagining things) that later drivers seem to do worse than earlier ones at deadlocking. I suspect they tightened the timings as they were squeezing faster performance out of the later drivers.

This one doesn't have ethernet, but my other overclocked 650 does, I've never noticed ethernet issues with it. It shipped 25MHz but I usually run it at 40MHz. I have a 66MHz PPC card in it that doesn't mind running at 80MHz.

Have you tried using a Nubus ethernet with the same chipset to see if the problem exists even then?

 

[eharmon]

This one doesn't have ethernet, but my other overclocked 650 does, I've never noticed ethernet issues with it. It shipped 25MHz but I usually run it at 40MHz. I have a 66MHz PPC card in it that doesn't mind running at 80MHz.

Have you tried using a Nubus ethernet with the same chipset to see if the problem exists even then?

The only NuBus card I have is an Ethernet NB (original, A/ROSE card), and it works fine. So, different driver (I'm running the later, non-A/ROSE driver), but it's definitely not the whole software stack being hosed.

If I switch to onboard ethernet it either hangs in a driver loop or deadlocks the minute anything tries to pass a packet. Same with every extension turned off except networking, so it's not a conflict. It's very odd.

 

[Phipli]

The only NuBus card I have is an Ethernet NB (original, A/ROSE card), and it works fine. So, different driver (I'm running the later, non-A/ROSE driver), but it's definitely not the whole software stack being hosed.

If I switch to onboard ethernet it either hangs in a driver loop or deadlocks the minute anything tries to pass a packet. Same with every extension turned off except networking, so it's not a conflict. It's very odd.

That is odd, I use ethernet on mine all the time. I have a socketted clock so swap speeds on it on a whim, and ethernet is my main way of moving files on and off it.

Have you tried a different transceiver?

 

[eharmon]

That is odd, I use ethernet on mine all the time. I have a socketted clock so swap speeds on it on a whim, and ethernet is my main way of moving files on and off it.

Have you tried a different transceiver?

It was all working until recently. I'm gonna keep fiddling so we'll see.

 

[cy384]

nice find! I don't think ethernet issues are common at 40 MHz, even on OS 8.1 mine is fine at 44 MHz (no idea if my ROM changes have any effect on ethernet, probably not).