See that Online Boggle Solver link on the right? Bet you’ve never clicked it!
I built the online solver back in 2006, mostly just for fun. In the six years since, I’ve wound up using it almost exclusively on my phone during real-life Boggle games. The iPhone didn’t exist when I built that page, and it is a device for which its fancy design (with hover effects and fixed position elements) is very ill-suited. Minimal designs tend to work better on Mobile, and I’ve just pushed out a new version for phones which is very minimal indeed.
The code behind this boggle solver differs from my other boggle code in that it’s designed for low latency rather than high throughput. This is precisely what you want from a web solver. You’re only finding the words on a single board, so time spent loading the word list dominates time spent finding words on the board. The current version uses a custom dictionary which can be MMAPped. This winds up being about 3x faster than reading a dictionary text file and building a Trie on every invocation.
I haven’t worked on Boggle in the past three years, so the state of the art in finding the highest-scoring boards is still where it was when I made my last blog post about breaking 3×3 Boggle. The interesting development then was that you could get something like a 1000x speedup by upper-bounding whole classes of boards rather than scoring each board individually. This is a big enough speed boost to make 3×3 Boggle “solvable” in a day on a desktop computer, but not enough to make 3×4 or 4×4 solvable, even on a big cluster of computers. I was optimistic that memoization could give some more big wins, but when those failed to materialize I got frustrated and stopped working on the project.