3 ms·
> how to value the sort of work that's more pragmatic than improving on a complexity bound but also more fundamental than just tuning some asm or C I'd bet laz
by jtolmar 5y ago
> how to value the sort of work that's more pragmatic than improving on a complexity bound but also more fundamental than just tuning some asm or C
I'd bet lazily slapping caches on things has saved more total time than all other computer optimizations combined.
- Retric 5y agoReally depends on the baseline, it’s easy to accidentally write something that would take billions of years to finish on real data but is fine for a mock-up. Caching often seems like the first optimization simply because systems tend to already be optimized enough to be useful.
- gwern 5y ago> I'd bet lazily slapping caches on things has saved more total time than all other computer optimizations combined. For compilers, "Proebsting's law" suggests the total gain from all compiler research to date has been ~3%/year or >3.3x total: https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.29.434&rep=rep1&type=pdf https://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.29...