Maybe we could help the situation along by building an architecture with absurd amounts of cache, given that cache is usually the FPS bottleneck of DF (or so I've heard.)
It's memory bandwidth in general, if I had to guess. Of course, memory bandwidth is the #1 killer of performance for almost everything. So, theoretically you could build a ginormous processor that had a 4GB register file (or cache, if you prefer a tad more sanity), but each one would probably cost $10,000 and would be hot enough to cook food on after it ran for a few seconds. Which of course is beyond the tolerance of any sane materials for an IC.
I also have doubts you could actually get the clock rate very high with such a large register file, so you might not get any benefit in the end...
Oh, but I'm rambling. I would be interested to see a cache profiling of DF. Some day I might actually try putting DF on a Linux computer and running Cachegrind on it (assuming it would run) and see for myself. If I had to guess, it probably does a reasonably good job on the cache performance.
DF just does so, so many calculations on so, so much data, it's the nature of the beast. Improvements in processor / memory architecture will help for sure, but parallelism is going to have to be exploited somewhere eventually to see the game truly scale.
This sort of thing should be best handled crowdsourced. We have the arena to test, the will to get a good result and lots of modders with the knowledge to properly tweak the raws.
If I had any say in all this, I'd tell Toady to let us do the boring part and focus on making dragons attack villages and properly grill the petty goblin raiders who happened to be there too.
Agreed. Some community data has been used in the past for things like materials, right? I'd like to see this more. If anyone does anything silly with the numbers, it can be reported on a case-by-case basis with the assumption that most of the data from the community would be good enough.