CCRL Discussion Board

I just did some calculations based on:

1) The time it took to solve tb5 slices on my 3.9 GHz i7-860
2) The speed difference between that and a single 5.0 GHz i7-980X with 6 cores (12 threads)
3) The possible gains from parallelizing with a twin Gulftown setup (12 cores, 24 threads)

Crunching all of the numbers, I still get 8 years 7 months to solve the entire set of 7-piece tablebases.
This includes 4 vs. 3, 5 vs. 2, and 6 vs. 1.

Seems like a very long time still, but much better than "decades" using any other technologies.

I have not fully spec'd out the rest of the system requirements (storage, RAM) but my guess is we're talking around $20,000 worth of hardware for this hypothetical "dream machine."

Thinking about hardware can be fun. Have you considered waiting 4 years before starting? Then you should be able to buy twice as fast hardware at half price and still finish 3.5 months earlier than if you buy now

koistinen wrote:Thinking about hardware can be fun. Have you considered waiting 4 years before starting? Then you should be able to buy twice as fast hardware at half price and still finish 3.5 months earlier than if you buy now

The hardware I mentioned will not be available for more than 4 years.

Consider this: In 2002, we had the first 3.0 Ghz computer. In 2010, the Intel Gulftown was released with 3.33 Ghz processing cores.

There has not been much progress in advancing the speed of the clock because there is a thermodynamic barrier as you get much faster than 3.0 GHz. More speed = more heat. What the computer manufacturers have done is improve the architecture, and offer more processing cores rather than faster clock speeds.

The 5.0 Ghz Gulftown system that I have deep freezes the CPU down to -40 degrees to remove the excess heat that is built up as a result of running the clock so fast. This single stage vapor change cooler is like a mini-freezer integrated into the computer's case. Unless Intel starts building refrigerators like this into their designs, I will always be able to make a system faster than the computer industry can provide for about 10 years.

Ed,

let's solve the KRPPKRP only. How long do you estimate that would take? I am ready to give some funds for this calculation, and I am sure there are others.

If we have still time, we should do the KBPPKBP, KNPPKNP, and lastly KQPPKQP.

ChessMonster wrote:Ed,

let's solve the KRPPKRP only. How long do you estimate that would take? I am ready to give some funds for this calculation, and I am sure there are others.

If we have still time, we should do the KBPPKBP, KNPPKNP, and lastly KQPPKQP.

Sorry to be an absentee for so long. Was busy improving my system to 5.27 GHz on the i7-980X, with its 12 threads cranking and solving some checkers endgames for me now



KRPPKRP would be the most interesting endgame in the 7-piece set, wouldn't it?

I will have to get an upper bound for the position count, and likewise for the dependencies, then try and do some math to see how long that one would take.