Satus of 7 men?

[jshriver]

Back when I was in the game 6men was hard but doable. Now honestly a high end gaming rig can do it.

What is the status of 7 men? I know it exist but for pay. Any ideas if the
Syzygy author is willing to make the code so? I know we have a lovely group of people with NICE hardware (myself included) who'd be willing to burn some CPU time on it.

I have a dual quad core (with HT thought I loath HT), and 32gigs ram on my main linux box I could burn for this. Plus 5-6 more puny systems.

Figured this forum is a great way to stay in contact, thread it out, and focus our efforts.

We all get 6 men... .then x does y, x1 does y1, etc.. we burn the first layer. Then I can make a website or torrent, so we can share, before we are all on level and work toward the 2nd layer. I have no idea how many layers, but I know KPPPKPPPP will be last lol.

Just poking curiosity. We need an open source 7 man TB. I can host it, at least seed in torrent and host for a couple months to get it out.

We just need a gen, time, CPU and people. Honestly WE HAVE THAT except for a gen.

[Evert]

This was discussed not so long ago. See the discussion here: http://www.talkchess.com/forum/viewtopic.php?t=62340

Bottom line: not so easy to do on consumer hardware.

[Jouni]

The benefit of 5 piece set is +5 ELO for top engines. May be +6 ELO for 6 piece set ? Just joking lets say 7 to 10 ELO? So practically no benefit for playing at all: but nice for endgame analysis and studies! 32 piece set will give perfect play e.g. +400 ELO or similar. But unfortunately we have not enough space in universum to store them. Is something like 10 piece upper limit in near future (30 years)?

[jhaglund2]

I would rather have GBytes of position learn, with format:

FEN - bm ply score

generate by the engine itself.

Crafty v25.4 would have this if it were up to me.

I did mentioned something like it before.

Take all the ECO openings,
At the end of each variation, learn each position searched, optional set max ply/time
Import FENs/EPDs analyze and create position learn from those.
Read PGN files and learn each move.
Load knowledge into memory, check for position to have occurred/ hit, return bm as search best move - root, etc

Then create an opening book for each ECO line. Let it learn.
Multiple opening books. Open the one that is played. 2GB+ of opening theory, plus 2GB position.lrn for example.

Way better than to try generate tables. I know my NVMe SSD reads 2,600 MB/s and writes 1,300 MB/s so there shouldn't be a problem accessing/ writing the knowledge with todays hardware.

[Dirt]

But unfortunately we have not enough space in universum to store them.

If we wanted and cared to we could build a 32 piece tablebase just using the material in this solar system. It would be a major undertaking, and I don't know if we could fill it before the Sun went red giant, so I don't think it would be worth the effort for a game.

[Dirt]

100 terabyte disk arrays are not common for consumers. Until they are I think the idea is premature.

[Dann Corbit]

100 terabyte disk arrays are not common for consumers. Until they are I think the idea is premature.

https://www.theregister.co.uk/2016/08/1 ... b_qlc_ssd/

[phhnguyen]

The benefit of 5 piece set is +5 ELO for top engines. May be +6 ELO for 6 piece set ? Just joking lets say 7 to 10 ELO? So practically no benefit for playing at all: but nice for endgame analysis and studies! 32 piece set will give perfect play e.g. +400 ELO or similar. But unfortunately we have not enough space in universum to store them. Is something like 10 piece upper limit in near future (30 years)?

Interesting info!

Do you have any number (ELO gains) for no top / "lazy" engines, (say, engines have no endgame code thus their endgames must be depended on tablebases)?