Interesting point of view! You could well be right.Kirill Kryukov wrote:We already have seen some long checkmates reported by Marc Bourzutschky. Yes they attracted a lot of attention. The result is that no one is working in this direction anymore, and we don't have any usable 7-men tablebases.
But the difference in size between DTZ and DTZ50 will not be close to a factor 1.5 at all. (Does anyone have statistics on the percentage of positions with DTZ/DTC>50 in typical 6-piece tables?)I agree that speed and size advantages are not the most important points, but they may be considerable when choosing between 1 TB and 1.5 TB to download.
Anyway, the decision between DTZ and DTZ50 can be made when the actual generation starts, unless the generation and/or compression code is for some reason limited in the range of values it can handle, in which case the choice is easy.
Btw, for the moment it seems more realistic to me to first aim for a (re)generation of all 6-piece tables using code that is open, efficient, uses sane indexing and ideally is prepared for 7-piece generation, together with a really good compressor that is specially adapted to tablebases.
Since the idea is to have 7-piece tables for practical play, should the format lend itself to access during search, or just at the root node? (And does anyone know what kind of storage devices will be available around 2025? )