Originally Posted by FOL
So alot of boards are missing these components? to make clock more stable?
I would not speak of "a lot of", as most of the customers do not experience graphics errors. However, since I had not been able to reproduce the bug here until today, there may be more "unfiltered" clocks out there than I'd expect from the range of boards that I have here, which all have the 14MHz clocks filteres with the ferrite/22pF combination.
I have just switched on the "cold" computer (that is: Board, power supply and accelerator are cold), and it works great starting off with an Alice temperature of 21 degrees Celsius, quickly working it's way up to over 46 degrees Celsius, where the computer starts to become unstable without the fix (measured with an Infrared thermometer). I consider this a successful test and rate the fix as "stable", so everyone out there who has graphics problems, please check your mainboards for E121R and E122R on the top side of the mainboard, as they're easily visible without too much hassle.
As always: Do not try to modify your board unless you do have the problems. If you apply the fix to a board with a different chipset combination, you might end up having a totally unstable computer - with or without accelerator.
Life would have been too good if this was the final problem (as I mentioned a few days ago). This weekend, someone on A1K.org pointed me to a problem that the FcCheck tool from WHDload forces: The computer freezes when you run this tool on an ACA1230. What it does is to make bus accesses with undefined/reserved function codes. Since they're undefined/reserved, my accelerator does not terminate these accesses. I find it highly questionable programming practise to use such function codes and expect the manufacturer of an accelerator to support them.
The closest thing to documentation I could think of is to give a "bus error" if such an undocumented/reserved access occurs. I have attached an image of the result. If someone who owns a "compatible" accelerator could post his result, that would be great (as it would really mark all reported problems of the ACA1230 as "solved").