SteveBagley wrote:exxos wrote:Also it seems to be running the CPU at 16mhz all the time, we never got that to work, there was some talk about it a few weeks back in this thread. Unless the issues were just with CPU <>MMU issues. If both are running at double speed then it could be possible the CPU could run 16mhz all the time. But then is also a big question mark on the E-clock on the CPU also.
If everything is sped up to 16MHz then the relation between DTACK, AS, and xDS lines will still be the same so everything should work fine, the problem comes when some parts are running at 16MHz and others at 8MHz because then the timing relationships expected by the CPU don't match what it gets…
As for the 6850s, I wouldn't be surprised if the 2MHz and 1MHz parts are identical and that each batch of chips are sampled to see what speeds they reliably work at, and then the whole batch labelled accordingly. Often you can find that several of the 1MHz parts are happy to run at 2MHz, but are just labelled as 1MHz.
Its possible some parts might overclock. Though I have ordered in some 2mhz ones to swap out just to be sure they are not going to be a problem. I needed to get some for the booster project anyway.
The 16mhz <> 8mhz clash is what worries me. As already known/tested the CPU will not run at 16mhz all the time. If the CPU can talk to RAM at 16mhz then thats fine, though where does GLUE fit into this. It looks like it must be running at 8mhz still. Though in my mind that is a GLUE 8mhz vs CPU & MMU @ 16mhz.
I was thinking of a more complex switch, to switch the MMU into 16mhz mode along with the CPU when the Shifter is not being accessed, then slowing the whole machine back down to default speeds. Though according to that file Arne posted, they are doing no clock switching like that. Just running the CPU at 16mhz all the time plus the MMU. The only fix is that DE part. Which would seem a little pointless if GLUE is running at stock speeds anyway. If all the clocks are still in sync with the master 32mhz then it wouldn't matter anyway.
I wonder if the DE thing is a fix for the "inverted" clock which happens earlier in the circuit. If all my assumptions were correct then it would pretty much render the entire mod pointless. A easier way would be just to buffer the 32mhz clock (delays wouldn't matter at that point) then feed the shifter and MMU with a new buffered 32mhz clock. Cut the 8mhz line from the MMU and connect it to the 4mhz line (so double clocked becomes back to 8mhz for the rest of the bus, GLUE etc). CPU gets 16mhz from the 8mhz MMU output, and that would be it. This would then suggested the CPU will not run at 16mhz all the time simple due to RAM access, and if RAM is double clocked, then the CPU is happy to run at 16mhz all the time. A lot of ifs/buts/assumptions with it all.....