6-men Syzygy from HDD and USB 3.0

[Dann Corbit]

Houdini 5.01 performs better in the famous Spassky vs Fischer endgame position if EGTB Probe Depth is adjusted.

I tried -10 and -5 on my brand new Samsung 960 EVO.

If syzygy is stored on M.2 SSD or equivalent, default Probe Depth is probably not correct setting for H5.1.

What depth worked best?
I have a big SSD also.

It's wrong to draw conclusions bases on few blitz games, but in my case -5 do seem reasonable.

Houdart argues that he don't like the speed penalty for probing tablebases, therby intelligent probing technique in Houdini 5.

For fast m.2 ssd, sweetspot is definitely between 0 and -10 somewhere.

Mine is probably not as fast, about 550 MB/Sec. Some of those m.2 gumsticks are absurdly fast (3 or 4 times faster than mine).

[mbabigian]

Kai, I have an Intel SSD SATA drive in one computer and a PCI card SSD drive in another. I also have 128GB & 256GB USB 3.0 thumb drives.

I have run 6 man Syzygy tablebases completely on SSD on both machines and compared the speed in late middle game positions against no tablebases at all. I see no significant difference in NPS with or without. In fact the IOPs are not high enough to believe that some of the new super fast SSD cards would make any difference over mine.

I have also run tests with all the tablebases on the 256GB USB 3.0 thumb drive as well as the WDL tables on the 128GB thumb drive and the DTZ tables on regular old spinning disk. Again, no significant difference between either configuration or pure SSD or no tablebases.

That said, I should also say that my machines have 32GB & 24GB or RAM respectively and I typically analyze with 8GB of hash leaving quite a bit free.

So from my prospective all the anxiety about accessing flash over USB 3.0 or splitting the WDL/DTZ files between flash and hard drive is overblown.

My two cents,
Mike
P.S. I also don't play bullet games. I'm talking about analysis (minute per move or more). So spin up to caching means nothing. Perhaps if you are playing game in 10 seconds the impact would be significant as nothing would be cached before the game is over.

[Laskos]

Kai, I have an Intel SSD SATA drive in one computer and a PCI card SSD drive in another. I also have 128GB & 256GB USB 3.0 thumb drives.

I have run 6 man Syzygy tablebases completely on SSD on both machines and compared the speed in late middle game positions against no tablebases at all. I see no significant difference in NPS with or without. In fact the IOPs are not high enough to believe that some of the new super fast SSD cards would make any difference over mine.

I have also run tests with all the tablebases on the 256GB USB 3.0 thumb drive as well as the WDL tables on the 128GB thumb drive and the DTZ tables on regular old spinning disk. Again, no significant difference between either configuration or pure SSD or no tablebases.

That said, I should also say that my machines have 32GB & 24GB or RAM respectively and I typically analyze with 8GB of hash leaving quite a bit free.

So from my prospective all the anxiety about accessing flash over USB 3.0 or splitting the WDL/DTZ files between flash and hard drive is overblown.

My two cents,
Mike
P.S. I also don't play bullet games. I'm talking about analysis (minute per move or more). So spin up to caching means nothing. Perhaps if you are playing game in 10 seconds the impact would be significant as nothing would be cached before the game is over.

I will measure NPS with long times on a set of positions and report. I only have 8 GB of RAM total.

[mbabigian]

I look forward to seeing your figures.

[Laskos]

I look forward to seeing your figures.

From 50 endgame positions with 30 seconds per position, I am experiencing a huge slowdown for small Probe Depths, to the point that using WDL+DTZ from HDD is practically rendered useless strength-wise. It seems time control doesn't matter too much, and choosing Probed Depth = 24 for HDD and = 12 for HDD+USB 3.0 was fine. Here are kNPS for these runs:

Code: Select all

---------------------------------------
Komodo 10.4  1024 MB Hash  4 threads  30s/position 50 poistions
---------------------------------------
No Syzygy:

kNPS:
9823
---------------------------------------
Syzygy-6 WDL+DTZ on HDD

Probe Depth     kNPS
 2              1077
12              2019
18              4318
24              7119
30              9273
----------------------------------------
Syzygy-6 DTZ on HDD; WDL on USB 3.0

Probe Depth     kNPS
 2              3418
 6              5812
12              8514
18              9779
-----------------------------------------

Keeping WDL on USB drive helps greatly, and probably gives boost Elo-wise, as I saw at ultra-fast controls.

[Jouni]

I took 100 longest 6 piece positions and check if engines can convert win with 5 piece syzygy (there can be 50 move draws). SF scored 79, Houdini 72 and Komodo 70. But Houdini 5.01 made illegal moves in some positions repeteably in Fritz GUI:
[D]RNN5/2k5/p7/8/8/8/K7/8 w - - 0 2

Here Houdini 5.01 wants to play Kxc8 I have seen Houdini doing illegal moves previously in Fritz GUI also.

[mbabigian]

Yep, WDL on HDD is awful. Flash + some memory for caching and you are in business.

I've never changed the probe depth from the defaults. Good info.

Mike

[Nordlandia]

Yep, WDL on HDD is awful. Flash + some memory for caching and you are in business.

I've never changed the probe depth from the defaults. Good info.

Mike

Houdini 5.01 performs better in certain engames if EGTB Probe Depth is adjusted.

Samsung 960 EVO M.2 SSD.

[tpoppins]

Houdini 5.01 made illegal moves in some positions repeteably in Fritz GUI:
[D]RNN5/2k5/p7/8/8/8/K7/8 w - - 0 2

Here Houdini 5.01 wants to play Kxc8 :?: :!: I have seen Houdini doing illegal moves previously in Fritz GUI also.

What GUI version is it? Under Deep Fritz 14 and Fritz 15 - both x64 - I see nothing but the single-move PV 2.Nxa6, with either 5- or 6-men Syzygy.

Houdini 5.01 performs better in certain engames if EGTB Probe Depth is adjusted.

A statement so vague as to be meaningless.

What are the measure and the extent of "better"?
What kind of endgames?
Adjusted to what values?

[Nordlandia]

Do latency caused by probing via USB 3.0 affect overall performance?

SSD connected via SATA III vs SSD connected via USB 3.0.

My initial observation is that USB 3.0 probing inflict conspicuous less TB Hits.

Probing 5-men via SSD trough USB 3.0 is fine. Probing 6-men is another story.