Houdini 6 endgame peculiarity?

[reflectionofpower]

Why does Houdini 6 drop way down in it's KNs when the board gets near the endgame. I have never seen this trait before in other engines. For instance:

[d]8/7p/p7/3kPpP1/PPp3P1/2R4r/3K4/8 b - - 0 41

[Jouni]

No slowdown, if You have tablebases cached to RAM or fast SSD.

[reflectionofpower]

I have 3-4-5 EGTB but these would not be affected as this position is more that what I have.

[Guenther]

I have 3-4-5 EGTB but these would not be affected as this position is more that what I have.

But those are of course used during search and you sure get already a lot of tb hits for this position soon.

[kasinp]

Quite the opposite here. As the position simplifies, NPs sky-rocket for H6, with tablebase hits going up rapidly too. In this position tablebases (even if only 5-men) are used extensively (into hunderds of thousands within seconds).

Something is not right in the setup you are using, I'm afraid.
Peter

[reflectionofpower]

Quite the opposite here. As the position simplifies, NPs sky-rocket for H6, with tablebase hits going up rapidly too. In this position tablebases (even if only 5-men) are used extensively (into hunderds of thousands within seconds).

Something is not right in the setup you are using, I'm afraid.
Peter

I have never had a problem with EGTB before until Houdini 6. I agree that nps should skyrocket but it is the opposite. IF every single engine I own does not do this except H6 then H6 has the problem especially if you say EGTB are even used in this position with all the other engines I have are NOT doing this. I am using them all in DF14 GUI. I even jacked the cache RAM to 512 just to see if it would do anything. It didn't do a thing.

[Jouni]

But have You syzygy or Nalimov?

[reflectionofpower]

But have You syzygy or Nalimov?

I use Nalimov.

[tpoppins]

I've seen NPS drop as the game approaches endgame with various engines, including Arasan and older SF. 5-men Nalimov is ~ 15 GB, so if you can spare 16 GB of RAM to cache the TBs, do it at Windoze startup:

Code: Select all

cd c:\nalimov
type *.* >nul

You can save above as a .bat file to the Desktop and run it as admin at startup.

If you can spare only a gig, switch to 5-men Syzygy. If not, 512 MB is enough to cache just the *.rtbw files which should account for 90% of the speed up. If you prefer Nalimov for its DTM, SF9 includes Ronald's "early mate" patch which greatly improves finding mates with Syzygy.

You can also to minimize NPS loss by raising EGTB Probe Depth in Houdini's UCI Options to a higher number -- experiment with various values until you're happy with the balance between NPS and search efficiency.

[reflectionofpower]

I've seen NPS drop as the game approaches endgame with various engines, including Arasan and older SF. 5-men Nalimov is ~ 15 GB, so if you can spare 16 GB of RAM to cache the TBs, do it at Windoze startup:

Code: Select all

cd c:\nalimov
type *.* >nul

You can save above as a .bat file to the Desktop and run it as admin at startup.

If you can spare only a gig, switch to 5-men Syzygy. If not, 512 MB is enough to cache just the *.rtbw files which should account for 90% of the speed up. If you prefer Nalimov for its DTM, SF9 includes Ronald's "early mate" patch which greatly improves finding mates with Syzygy.

You can also to minimize NPS loss by raising EGTB Probe Depth in Houdini's UCI Options to a higher number -- experiment with various values until you're happy with the balance between NPS and search efficiency.

Ahh, here is the answer. EGTB probe Depth. I put it at 25 and it screams like a banshee now,thanks. My hats off to "T. Poppins".