fresnault
commented
9 years ago Because it's not possible to make a lot of reservation on Igrida, I run 3 sort of jobs :
- Small job during the day (9h-17h)
- Medium job during the night (17h-5h)
- Long job during the week and week-end
| Name | Number of jobs | Duration | Number of FEN to analyse per job | Total of FEN to analyse | Total worktime |
|---|---|---|---|---|---|
| Short Job | 10 000 | 8h | 6 000 | 60 000 000 | 80 000h |
| Medium Job | 9 000 | 12h | 10 000 | 90 000 000 | 108 000h |
| Long Job | 4 000 | 64h | 30 000 | 120 000 000 | 256 000h |
| Total | 23 000 | - | 270 000 000 | 444 000h |
We can see that it takes 444k hours to analyse our 270 millions FEN. To give you an idea, that represents 18 500 days (50 years) on a unique machine.
Our 270 millions FEN takes only 15.5 Go memory before analysis. Yet, each analysis has 3 ko logs (~3000 characters per log). So, we expect about 800 Go logs.
In average, Stockfish calculates 7 millions combinaisons at depth 20. So, we can think Igrida has, at total, calculates more than 2e15 nodes (~2 000 000 000 000 000).
Can we have key numbers from the analyse ?