exxos blog - random goings on

[JezC]

I need to read this in greater detail...but it was running for pretty much the whole of CL4 (probably 4+ hours) with only a single manual restart (while helping CiH sort HDDriver for his display Falcon), so no crashes or lock-ups then.

Maybe my struggle to get it out of its case and/or the travel back has disturbed something?

PS - thanks for your efforts on this, I was primarily intending to help you with the clock patches not to completely derail all your investigation!



Feel free to put it aside for a bit and make real progress with your other projects - I have plenty of other stuff to burn my time until December at the earliest (and more realistically February or later)

[exxos]

@JezC I think part of the problem is it seems unhappy with the DFB1X... but I think the board likely flexed and broke the SDMA clock.. The SDMA has been changed at some point and the soldering isn't great. The clock pin didn't even look soldered down.

It ran 2 hours last night. I've just started it up again. I think the SDMA clock needs to be above 4v to work. So that's something learned. The V3 barely hits it. I need to change it for a V4 as that has higher output voltages.

The original clock patch when I put it back, the machine was still unstable.. So I don't know how it even worked in the first place. As changing the big cap helped, I think you had one of my PSUs in there? So it could be power related as I'm testing with a original PSU. Regulation not as good so a original PSU could have tipped it all over the edge I suppose.

[JezC]

It had one of your Falcon PSU replacement units, yes.
I've only used it with that for the last few years after reading the horror stories of high voltages from failing original PSUs killing custom chips & RAM etc.

It put up a valiant fight before I finally eased it out of the case, so that and/or vibration could easily be involved I guess.

TBH I don't remember having many problems with it once your PSU was fitted... either with or without the DFB1X (bar getting the essentials like NVDI installed so it booted ok in accelerated form) so it might have been 'stable' but working on the edge since then.

[exxos]

Choices choices...

IMG_3897.JPG (263.41 KiB) Viewed 188 times

IMG_3898.JPG (257.35 KiB) Viewed 188 times

IMG_3900.JPG (379.88 KiB) Viewed 188 times

So the 100pF on the FPU doesn't seem happy on this board.

IMG_3899.JPG (275.68 KiB) Viewed 188 times

I've just removed it and retesting again as GB6 locked up after a couple of runs. So currently testing again..


So the raw V4 SDMA output is:

v4.png (46.18 KiB) Viewed 188 times

Then with 62R series resistance.

v4_62r.png (44.3 KiB) Viewed 188 times

Then with the 100pF on the FPU.

v4_62r_100.png (43.63 KiB) Viewed 188 times

So without the 100pF there's more ringing. But it may not be stable with that in place.

In my install I had 47pF on the SDMA clock as well. Not fitted that yet. I may try that next, or I may try a higher value resistor like 100R. Eitherway it's a lot less ringing than the original mod had.

Will leave it running as-is for a couple hours, likely will give the DFB1X a try before trying other SDMA tweaks.

EDIT:

GB6 passed 15 full passes. So I put 47pF on the SDMA clock and doing more tests. (100pF still not fitted)

The clock looks like this now.

v462r47p.png (116.2 KiB) Viewed 184 times

[exxos]

So that didn't work. 47pF now removed.

I guess its possible while the ringing can be solved, maybe the slight delay it causes on the SDMA clock causes it to become unstable

If its stable without the 47pF, I may see if I can find my PLL prototype board, as that compensates for the SDMA clock delay. Then I could retry the 47pF to see if clock sync is a factor or not...

and its locked up again

I will connect the SDMA clock to the combel clock then it will be a few ns ahead of the combel clock, but by the time its got to the SDMA it ends up 4ns slower anyway, but 2 buffers less it should basically all be in sync then...

EDIT:

Nope So will have to put back the V3 and rerun tests again as that one uses inverters not buffers.. This means combel is then out of sync with everything else..

EDIT2:

Nope V3 isn't working again now.. I don't get it. Was working last night. Maybe I am barking up the wrong tree and there's some other odd fault on this board somewhere.

EDIT3:

I've removed the 62R from the SDMA line as that wasn't there yesterday.. running tests again now.. The SDMA clock doesn't look great. High of 3.80V or thereabouts.

I did design a V6 clock patch a few months ago. That one is a mix between V3 and V4. But it has the option for inverting and noninverting .

Assuming this is a SDMA clock fault, the V6 would up the voltage to the SDMA clock.. If it works worse.. then maybe the SDMA doesn't like a higher voltage...

Will have to see how V3 works as another baseline test now.

I suppose - I could hack on a ~6V regulator on the V3 to up the voltage output a bit to see if that helps or not first...

[exxos]

The V3 has been running for about an hour. So that is good because I basically have a solid base line.

Oddly even just 62R on the SDMA clock made it unstable.

The difference before and after the resist is hardly anything other than a marginal slower rise and fall time and a little less ringing.

I don't believe it is a clock sync issue, but I still cannot totally rule it out at this point.

I will get my bench power supply and power the clock patch from that so I can alter the voltage and see if this has any effect.

[exxos]

My jank setup..

I know 5.10v works. So trying 4.50v now...

SDMA clock gets 3.12v max according to the scope..

20251014_180313.jpg (266.87 KiB) Viewed 152 times

Now to wait a hour

[exxos]

No problems with 4.50v.

Now upped to 5.80v and sdma clock is at 4.40v... Now another hour wait...




My current thought is that maybe the clock trace actually needs high current in order for it to not pick up interference. That when a series series resistance is present, it must allow more interference. It makes sense that that could happen, but I haven't really detected any interference on the clock of any sort of magnitude, so I don't see how things are fitting together currently.

If the current test works then I have mostly ruled out clock sync and clock voltage issues. The problem is higher current also means higher ringing, you cannot really have it both ways.

I may try the second part of the V4 clock patch which buffers the SDMA clock directly, as that may be the best of both worlds, even though it does actually delay the clock slightly more.

The weird thing is though that the V4 outputs a higher current than the V3 so based on my current hypothesis you would think that it would work better but the problem is the V4 is a non-inverting buffer which could also be a factor.

Atari themselves to say there was interference on the clock signal, but maybe they got around the problem by using an inverter. Which shifts some timings about and reduces the noise picked up by the SDMA trace... Maybe I could out the V4 back and try a new wire to see what happens..

[exxos]

5.80v test resulted in even more weirdness. It reset a couple of times and then refused to boot up at all. It was like the reset button was being held down continuously. I had no monitor sync and the CPU clock was just stuck in 8 megahertz.

I put the DFB1x in and the white LED was lit brilliantly indicating that the machine was stuck in reset pressing the reset button did not change anything. After about 30 seconds, the DFB1X kicked in and the machine booted up, so I removed the accelerator and the machine booted up and got about halfway through running GB6 and then reset again. I had the clock patch running at 5 volts so I don't believe that is an issue.

So now I don't know what's causing this intermittent resetting issue.

[exxos]

So the reset button is working, but now I am only seeing 0-1V instead of 0.5V..

Didnt you @mikro have issues with the reset circuit at some point ?

...and where the hell is C7 ?

The back end of D1 to GND, short when reset button pressed (so fine) resistance on my meter jumps to about 500R and stays there.. it should keep going up.. So I wonder if C7 has died and shorting out the reset circuit ..

Capture.PNG (286.84 KiB) Viewed 192 times

I think this is C7 going by is connected to D1, but no change

IMG_3903.JPG (294.32 KiB) Viewed 185 times