The New Sudoku Players' Forum

[Serg]

< Withdrown >

[dobrichev]

The recent estimations for {-3,+3} search over the known 17s show that about one year is needed on a 32 core 3.2 GHz Ivy Bridge CPU.
A single thread {+3} on the alphabetically first 17-3 subgrid takes 28.327 seconds = 310 days on 32 cores. The 10 alphabetically last subgrids are processed for 335.182 seconds = 367 days on 32 cores.
This is about as twice as faster than the estimations done in the last years but twice slower than the estimation in my previous post.

[champagne]

The recent estimations for {-3,+3} search over the known 17s show that about one year is needed on a 32 core 3.2 GHz Ivy Bridge CPU.
A single thread {+3} on the alphabetically first 17-3 subgrid takes 28.327 seconds = 310 days on 32 cores. The 10 alphabetically last subgrids are processed for 335.182 seconds = 367 days on 32 cores.
This is about as twice as faster than the estimations done in the last years but twice slower than the estimation in my previous post.

Hi Mladen,

just to catch clearly what id done
{-3,+3] means

-kill 3 clues among the 17 known
-add 3 clues taken in the 54 unknown

all that in blind mode (any combination)

is it right

[dobrichev]

Right.

[dobrichev]

Some {+3} timings measured on 2 x Intel(R) Xeon(R) CPU E5-2680 0 @ 2.70GHz (Sandy Bridge, total 32 threads on 2 x 8 physical cores).

The source data are 30 302 691 essentially different 14-given subgrids at {-3} to the known 49157 17s, lexicographically ordered and partitioned to 304 chunks of size 100 000.

Chunk 000 (100 000 subgrids)
CPU time 4622958 seconds (= 54 days)
Execution time ~ 48 hours
Puzzles produced 236244
Essentially different puzzles produced 24720

Chunks 001 to 301
On hold, waiting for quantum/gravitational/other computer.

Chunk 302 (100 000 subgrids)
CPU time 4934273 seconds (= 57 days)
Execution time ~ 53 hours
Puzzles produced 131158
Essentially different puzzles produced 4310

Chunk 303 (2691 subgrids)
CPU time 145830 seconds (= 2 days)
Execution time ~ 1 hour
Puzzles produced 3434
Essentially different puzzles produced 1435

For better readability the number of new 17-clue puzzles found along with some other zeroes are not shown.

[champagne]

Hi Mladen,

I don't catch clearly what is the final goal of that process.

Is it a tentative proof that you can generate all known 17's out of the 14's patterns ?

what else ??

[dobrichev]

27.7% of {-3,+3} is done with no new 17s found.

[blue]

Hi Mladen,

27.7% of {-3,+3} is done with no new 17s found.

Did you ever finish the {-3,+3} search ?

[dobrichev]

Yes. Months after your latest 17 findings. The {-3,+3} search should found some of them, but they were already involved in the mill.

[blue]

Thanks for the update.

For better readability the number of new 17-clue puzzles found along with some other zeroes are not shown.

I didn't catch that the first time around.

[dobrichev]

Actually I've reported the {-3,+3} finalization here.

[blue]

Yes, indeed.
After seeing it again, I remember reading the {-3,+3} thread.

Apologies.
I was looking for this post, and found myself here.