Back in late March, my watercooled system developed an odd problem: it would not post unless I warmed it up to about 75F. I run the thing for months on end without powering off, so this wasn't a major problem - just odd.
I finally got around to disassembling the system and spotted the problem:
Not so good. The VRM waterblock retention mechanism damaged two of the CPU MOSFET surface mount capacitors.
I replaced the motherboard and reassembled the machine, but it hard locked while I was adjusting BIOS settings. Upon disassembly, I found leaked coolant. Fortunately, I am using distilled water, iodine, and inert dye, so nothing was damaged.
I assembled the loop separately:
Running it for a few minutes revealed the source of the leak:
The end of this tube has been cinched one too many times. Easily fixed.
Glenn Fowler of AT&T Labs
wrote an excellent Sudoku application - Sudocoup
- which I have envied for some time. As I developed my own Sudoku solver, I compared its execution time for various puzzles to that of Sudocoup.
In April, Sudocoup could solve 9x9 puzzles that mine could not. By June, my solver could do any puzzle up to 36x36 - but far slower than Sudocoup. In July, I implemented a raft of additional constraint logic reduction routines that allowed solving 64x64 puzzles in a few minutes - as compared to the seconds required by Sudocoup. 144x144 remained completely out of reach.
However, I finished implementing single-node chain reduction for my solver, which I named Arcterik Sudoku. Arcterik Sudoku solves this 256x256
puzzle in 0.635
seconds, whereas Sudocoup requires 3 minutes, 26 seconds
So, I've hit the faster than Sudocoup
milestone that seemed unreachable for so long. My ultimate goal for this project, aside from sharpening my graph theory and learning some new things, is to handle 65536x65536 puzzles in reasonable time.
16x16 in the titlebar refers to block size - 16*16=256, the width of the puzzle