jwd wrote: 05 Jul 2024 10:46
What I will say is that even though the phantom cleaned it up (SDMA) nicely there was still some 'artefacts' present in the signal (compared to the super clean output from the powerup2) and if I wanted to put a further speeder (DBFX or Nemesis etc.) then I would probably need a cleaner signal to SDMA as it would all be put under tighter tolerances and therefore I will probably need (in my case) the V4 patch.
Just a small note here in case of future readers etc.. While the V4 does give the best signals & fixes to date, I'm not suggesting people use it with any sort of bus accelerator like the Phantom etc for many reasons.
The problem is, Phantom etc is pushing bus timings so tightly that even something like 1ns difference on the clock can make or break the system. Originally I thought it was just a simple matter of a delay on the clock, hence using the non-LS buffers on the original Phantom clock patch, so I imitated that delay with variations, which was why myself and
@Steve tried a delay board which did 10ns steps. However I did read somewhere that I think the original designer of the Phantom, changed the brand of buffer chips at some point and it caused problems. That seems to be how tight the timings are. In which case it is more of a fluke if some hardware combination works or not. The V2/V3 mimicked the Phantom delay, I think people had some success with that with bus speeders, but again I think it is more of a fluke of hardware timings as to if it would actually work or not. Unfortunately when you push the bus speed to the limits, you're going to get tight timings and with enormous amount of tolerances in the hardware across all machines, it further complicates issues.
So because the timings on the bus accelerators is pretty machine dependent and indeed such variances on tolerances, there is just no way to build a clock patch which is going to work in all machines with bus speeders. You are fighting minute tolerances in all the timings on all chips. Other than trial and error, your never going to get a "plug and play" solution to this problem.
The Falcon itself may not always boost to 25mhz (or whatever) reliably to start with. So developing a clock patch to correct such problems is basically impossible because it simply cannot be done. So in terms of the Phantom, people should probably use the Phantom clock patch exactly how it's creator intended it to be, and use the exact brands of chips as was used originally. Though even doing that, tolerances on the buffer chip itself or tolerances on the Falcon can still break such a system. I'm sure people know that overclocking is a bit of a roulette gamble anyway.
So in light of all these problems, I pretty much abandoned the bus accelerator aspect altogether. I would just rather keep out of it all, because I don't want to get drawn into all the chaos it would endeavour when people are trying to overclocked their machines and things don't work as expected.
Saying that some falcons with no clock patch whatsoever can run speeders/accelerators and the JOY demo absolutely fine, each Falcon must be considered unique and what works for one Falcon might not work for another. Another example is the capacitor in the video circuit that is the wrong way round, voltages need to be measured and there are variances in the circuit on the board that aren't in the schematics, mine is correctly orientated from factory but vast majority aren't again need to measure and check your own bird.
Yep. I don't think anyone has really looked into such specific details before. I think my original Falcon does not have a clock patch if I remember rightly. I definitely used it with a hard drive and don't recall having any issues as such. But mostly I was using STOS where I could easily blame buggy code for crashes, but its possible there is simply more to that now more is known these days.
Though it is still pretty weird why some falcons would seem to work without a clock patch at all. I'm not saying they are 100% stable. The odd crash every several hours can easily go unnoticed for example. How the resistors are wired to the single clock is just inherently bad start with. It is possible some machines may work fine on hot or cold days but not the other for example. Same with bus speeders.
Maybe there was some very early falcons which did work reasonably well, maybe Atari changed the supplier of chips or did something layout wise on the motherboard, even a totally separate production run could be enough upset things. But obviously Atari recognised the issues, hence doing the modification to pretty much every Falcon since. We could simply put it down to a fluke that some machines just inherently work while the majority don't. Just moving a trace couple of millimetres can be enough to cause total chaos, I have witnessed that first-hand.
TL;DR;
The V4 solves all currently known clock issues to date, but it should only be used on a stock falcon. People who want to experiment with bus speeders are "on their own".