First post, by BitWrangler
- Rank
- l33t++
Hi folks,
I've really got too much to "keep everything" long term, and there's unnecessary duplication, start digging into details too much and everything is a unique, precious snowflake. Then you're trying to judge things on intangibles. So I was looking for some kind of rule or structure that made sense to me, that I could keep myself to, and I think I've got it now, at least for a "first pass optimisation" as time goes on I might cut more dead wood than this. That depends on whether things feel too similar cross architecture or cross generation.
So the basics are, that I've known for many years that you don't really feel a performance difference on a system until you go at least 20% faster, and even then you're probably only noticing the edge cases, things where it wasn't quite fast enough before. Really you need 50% to double before you get the "Ohhh yeahhh!!" feeling. So for things like that, where there needs to be an amount of separation for a real difference, the concept of preferred numbers exists, the E3 series https://en.wikipedia.org/wiki/E_series_of_pre … rred_numbers#E3 has a calculated tolerance of 36.6% which seems in the ballpark of the kind of difference I want between systems, so they actually feel different to use. But 1, 2.2, 4.7 isn't actually all that brain friendly for rough reckoning so rounding up to the 1-2-5 series https://en.wikipedia.org/wiki/Preferred_numbe … 2%80%935_series seems good enough...
In essence then, for each architecture, I would have the 1x fast, the release speed, the 2x as fast, close to 2x release speed, and the 5x as fast if possible, top end of the speed range. But that's not real rigid and the plus or minus 36ish percent gets you in the same ballpark.
For example, for 486s, I'd go 25 or 33, then any DX2, then DX4 120 or DX5 133.
If I consider; XT class, 286, 386, 486, socket 5/7, PII/III, P4, K7/462, AMD 64, and conroe/Core2 as the generations or architecture divisions, then that's still some 30 machines or board combos, but as I said, I've already got duplication, so it should be a reduction. Also as I mentioned, may just be a "first pass" since there's definite areas there where the performance envelopes overlap between competing architectures. I might wanna experience them for a while before deciding. I also don't know if I'm allowing myself a couple of special cases, like a dual CPU, and the exact spec of late 1995 I had.
So any thoughts on your curation philosophy? ... other than the Pokemon one... gotta catch 'em all.
Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.
