Originally Posted by Loedown
Not appearing to be rude Jens, but if hardware is properly made in the first place with reliable firmware why would it fail?
No worries - a question like that doesn't let you appear rude, just inexperienced. The more complex a product becomes, the higher the probabilities that there's a special config where it fails. Not everything can be covered in lab tests, especially not combinations with other, either unknown or undocumented, maybe even non-standard products.
The whole trick is to keep complexity down. Allowing a memory socket instead of a fixed amount of memory would add the problem of covering all the different organisations that memory modules come in (easily five different configs only on SD-Ram 168-pin DIMMs, multiplied by the different speed grades). With a fixed amount of memory and all accelerators tested to work with this memory, I can reach zero customer support cases due to memory incompatibility.
A PCI slot allows so many different cards and requires so many different drivers that it's safe to say, you will never finish development of such a product. There's always one more driver to write. That's why I said: no PCI from my side. I'm already worried about timing specialties with "whatever you can connect between A1200 board and accelerator", but that's a standardized interface that can be tweaked with the ACAtune tool. The board cold-starts with a safe config, and if a tuning option does not work with a specific expansion, you just switch it off (and please don't ask me to make it faster anyway - I can't!).
Remember that this was meant to be a memory expansion. Faster speed is an added bonus. With all the added features of the ACAtune tool, we've exceeded the complexity that we aimed at by far already. We welcome and consider every suggestion, but if it adds too much complexity, it will affect our target quality, which is why we *have* to say no to so many suggestions. We can't argue with reality.