exxos's Atari, Amiga & retro forum

reading this i remember same issues here on german side, no chances to run a card with @25Mhz using 64MB SDRAM expansion.. may be same reason... :?

exxos wrote: 26 Nov 2018 22:20

terriblefire wrote: 26 Nov 2018 22:12 I agree. I've just found the CD32 very picky on this. Its actually been quite easy to get 50Mhz working. 25Mhz is not happy at all. I think its because 14.8Mhz and 25Mhz are too close together. You cant fit 2 fast cycles inside one slow cycle. Perhaps I should just fit an 80Mhz crystal instead and go with 40Mhz and be done with it.

EDIT: Amiga's dont do ARB and slow down BTW. They cycle overlap and crash. its a real PITA.

If your trying to fit 2 fast cycles inside a slower one, thats probably why slower speeds are not working. I half recall you talking to me about this before ? I think you was trying to run fast-ram cycles at the same time as "MB" cycles were happening ?

Even so, if clocks are out of sync its complicating matters as that fast clock timing is going to be a bit "hit and miss" in the lower clock domains timing.

EDIT: You probably would need a fast domain clock, I would guess 2.5 or 3x faster than the 14MHz clock to do 2 fast cycles inside a slower one. Probably like 42MHz would be needed to complete 2 full clocks reliable inside a 14MHz clocked cycle.

Yeah. 21Mhz was the lower limit for the TF530 with how I wired it up. Makes sense that this one is 42Mhz with everything x2.

About to try putting DSACK[1:0] through a 70ns delay to see if thats enough.

Yeah that didn’t work. I think 25mhz mode may be a non starter. I wonder if the 28mhz motherhood signal is on the riser somewhere. That would be easy.

You might be making things worse adding a delay. I mean 25MHz is 40ns cycle time. So you can be 40ns faster or slower than whatever you are timing it to. So adding 70ns would mean you will be 40ns+70ns = 110ns later than your expecting depending on where the cycles happen to align up.

exxos wrote: 27 Nov 2018 08:35 You might be making things worse adding a delay. I mean 25MHz is 40ns cycle time. So you can be 40ns faster or slower than whatever you are timing it to. So adding 70ns would mean you will be 40ns+70ns = 110ns later than your expecting depending on where the cycles happen to align up.

Indeed. I may just give be up and use the 28mhz signal for slow running. The CPLD is already very full with the SDRAM code.

Would it be possible to take the 14 mhz cpu clock @ pin 99 and multiply it thus eliminating the need for onboard crystal? This way everything runs in sync. You get 28/42 mhz speeds for the 68030 kinda like the M-Tec accelerator on the A1200.

EDIT: looks like you are already thinking along this line. NVM

I was but when Chris and I tried a clock multiplier there was an issue.. can’t remember that it was now.

To be clear this issue is completely unrelated to the SDRAM. It’s just the usual nonsense you have to deal with every time you change the clock relationships.

terriblefire wrote: 27 Nov 2018 13:41 I was but when Chris and I tried a clock multiplier there was an issue.. can’t remember that it was now.

The PLL multiplier I tried was responding to noise and upsetting the PLL frequency. So while it worked 99% of the time just fine, it was causing random crashes as the "locked on frequency" changed... IMHO that kinda defeats the object of a PLL if it adjusts to the input frequency in real time :roll: A PLL should lock onto a signal and bloody well stop there, but I haven't found one which actually does that.. In the end I gave up screwing around with them. I probably documented all this "*somewhere* .

EDIT:

Oh and the x2 settings wasn't even reliable, it would end up at like x5 half the time. Probably if enough time and effort went into it, it could be made to work properly. But they made my "never use again list", not worth the trouble.

terriblefire wrote: 27 Nov 2018 13:41 I was but when Chris and I tried a clock multiplier there was an issue.. can’t remember that it was now.

This one does this task excellent:
https://www.digikey.com/product-detail/ ... ND/2005327
Have it on a 020 accelerator card