DFB1X with Nemesis / clockpatch

[ctirad]

Yes, 50MHz system clock + DSP/25Mhz BUS.
The machine boots with 25MHz BUS enabled, but sooner or later it freezes completely including mouse cursor. I want to try adjusting the crossdomain synchronization code. Also, Claude AI said the current code can result in hazards even at 16Mhz.

[exxos]

Have you had a logic analyser on all this? A stock Falcon may tolerate overclocking the bus to 25MHz, but the DFB1/X adds PLD delays to the mix, probably around 10ns. That sits on top of timing that is already being pushed.

Was nemesis a solid bus boost or did that do clock switching itself between 16mhz and 25Mhz?

Before rewriting the clock domain crossing, try it with acceleration disabled via the jumper. If it still freezes at 25MHz bus, the mux isn’t switching at all, so the clock switching code can’t be the culprit. The “hazard at 16MHz” is a known trade-off BW made on purpose for switching speed, and if it was the cause it would be falling over at 16MHz too.

dml’s result of solid at 20MHz and flaky at 24MHz once it warmed up smells far more like timing margin than logic. A scope on the clock input to the booster and an LA on XAS/DSACK against the bus clock would tell us more than any amount of code reading.

[ctirad]

I haven't used the LA yet.

Phantom switches the main clock via software and once switched, the speed-up is permanent and rock-stable.

It is also perfectly stable with DFB1x disabled.

It also works without any issues with the CT63.

Could you make the DFB1x source code available to me, please? I'd like to try some modifications while I have the time and inclination

[exxos]

Sorry, but the DFB1X source isn’t available. Keeping it closed was part of the agreement I made with BW when I took the DFB1X into production. It stays closed while the boards are on active sale, or until that agreement runs out. It’s not about being awkward. It’s simply there to stop the design being cloned while it’s still being sold. When the time comes, it’ll be released, but not before.

[Badwolf]

Could you make the DFB1x source code available to me, please? I'd like to try some modifications while I have the time and inclination

If you were so inclined, you could build a DFB1 from the DFB1X parts, but I appreciate that's a proper PITA!

If anyone comes up with a better clock switch solution, please submit a pull request. I spent a few nights on it a couple of months ago and didn't find anything better than what's there now (Acetracker really hates my clock switching and I was workign on the DSP/GAL issue).

My suspicion is that exxos is correct and the faster clock has used up what margin there was in the design. Simply inverting the system clock may solve the problem by, in effect, adding in a whole bucketload of new margin.

If you have a clock patch installed there may be a quite simple mod option to try without needing to change firmware.

BW

[exxos]

I'm not convinced it's even a clock switching problem. If it was the clock glitching, surely with faster clocks the glitches would be less anyway... It needs to be scoped and proven before anything else.

Id bet it be more something like needing a stiffer pullup on DTACK or something along those lines which may help with timings. Or a proper PLL clock patch to make the PLD's 10ns "vanish" which would give everything more timing margins. Not saying it is that, but thats probably where I would start.



Figures on a example as I'm too lazy to figure it out myself .

Capture.PNG (92.64 KiB) Viewed 25 times

But needs to be measured on real hardware before doing anything else IMO.

[dml]

(reading the end of the Claude dump...)

I didn't report 'temperature drift' but unexpected instability after reinstalling the fan to the board. I didn't reassemble the machine and the fan wasn't fanning anything - except maybe my face

So I read that as a noise/proximity problem - or some other disturbance I couldn't identify after that change.

Before the fan was soldered/bolted back onto the board, it did run at 24mhz and passed lots of tests with no bombs or freezing. I was 'a bit surprised' but it seemed like a good result.

Fitting a buffer (phantom?) board made it super-crashy immediately so I removed it, put the dividers back and it was stable again.

I fitted the fan, and it went slightly-unstable and stayed that way. I couldn't run tests without crashes. It was happy though at a slightly reduced 22MHz.

[Badwolf]

I fitted the fan, and it went slightly-unstable and stayed that way. I couldn't run tests without crashes. It was happy though at a slightly reduced 22MHz.

Is the fan still on the 12V rail? I know it a common mod to put it on the 5V rail, but that seems like a bad idea to me.

BW

[dml]

Is the fan still on the 12V rail? I know it a common mod to put it on the 5V rail, but that seems like a bad idea to me.

I just restored the connections that existed already - the fan wires were cut to clean the board properly, later restored with heatshrink. So it would be on the 12v rail.