Factors in search... What is important?

Discussion of chess software programming and technical issues.

Moderator: Ras

Dann Corbit
Posts: 12845
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Factors in search... What is important?

Post by Dann Corbit »

I have a big pile of positions that Ezekiel is analyzing right now. This is the first position in the pile:
[d]r4rk1/1bpqbppp/p1np1n2/1p2p3/4P3/PB1P1N1P/1PP2PP1/RNBQR1K1 w - -

Here is the output of AsmFish, to ply 37 (my normal analysis depth):

Code: Select all

Analysis of /media/danniel/sas_orig/chess/n37c.epd   
21-8-17 09:12:42 Level: 37 Ply
Analyzing engine: AsmFishL_2017-08-19_base

 1                      
    Avoid move: 
    Best move (AsmFishL_2017-08-19_base): Nb1-c3
    Not found in: 10:29
     2/2	00:01	          8k	273k	-0.15	Nc3 Nd4
     3/3	00:01	          9k	303k	-0.04	Bg5 Na5 Nbd2 h6
     4/4	00:01	         10k	331k	+0.18	Bd2 a5 Nc3 Nd4
     5/5	00:01	         10k	341k	+0.18	Bd2 a5 Nc3 Nd4 Ba2
     6/7	00:01	         40k	914k	+0.05	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5
     7/7	00:01	         53k	1,104k	+0.44	Be3 Na5 Ba2 h6 Nc3 Nc6 Nd5 a5
     8/8	00:01	         57k	1,142k	+0.44	Be3 Na5 Ba2 h6 Nc3 Nc6 Nd5 a5
     9/10	00:01	        186k	2,117k	+0.34	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Nf4 Rfc8 Bd2
    10/13	00:01	        508k	3,576k	+0.54	Nc3 h6 Bd2 Rad8 Nd5 a5 Nxe7+ Qxe7 a4 b4 c3
    11/16	00:01	      1,050k	5,555k	+0.40	Nc3 Rae8 Bd2 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Qc8 Ba2 h6 Nf5 d5 Qf3 Qd7 exd5
    12/20	00:01	      1,873k	8,178k	+0.35	Nc3 Na5 Ba2 c5 Nd5 Nxd5 Bxd5 Nc6 Bd2 h6 a4 Nd4 Bxb7 Qxb7
    13/19	00:01	      2,657k	10,219k	+0.40	Nc3 Na5 Ba2 c5 Bd2 Nc6 Rb1 Rad8 b4 cxb4 axb4 Nd4 Nd5 Nxd5 Bxd5
    14/21	00:01	      7,008k	16,529k	+0.40	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Bd2 Rac8 c3 dxc3 Nxc3 a5 a4 c4 dxc4 bxc4
    15/22	00:02	     20,027k	26,386k	+0.37	c3 Na5 Bc2 c5 d4 Rac8 d5 Bd8 Nbd2 Bb6 Nf1 Nc4 Ne3 h6 Nf5
    16/25	00:02	     26,373k	28,729k	+0.27	Nbd2 Nd8 c3 c5 d4 exd4 cxd4 Ne6 d5 Nf4 Nf1 N6h5 Bc2 g6 Ne3 Bf6 Ng4 Bg7 Bd2 f5
    17/29	00:02	     29,146k	29,500k	+0.35	Nbd2 h6 c3 Na5 Bc2 c5 b4 Nc6 d4 exd4 cxd4 cxd4 Nb3 d5 e5 Ne4 Bb2 Qc7 Nbxd4 Nxe5 Nxe5 Qxe5 f3
    18/26	00:02	     45,735k	27,785k	+0.27	Nbd2 h6 Nf1 Nd8 Ne3 c5 Nf5 Ne6 c3 Rfc8 a4 Bf8 Bc2 Nf4 Bxf4 exf4 d4 c4 Nd2
    19/27	00:03	     49,152k	28,428k	+0.30	Nbd2 h6 c3 Na5 Bc2 c5 b4 Nc6 Bb2 Rfc8 d4 exd4 cxd4 cxd4 Nxd4 Nxd4 Bxd4 Nh5 Nf3 Nf4 Bb3 Ne6
    20/31	00:03	     80,223k	32,076k	+0.33	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Bd2 Rfc8 c3 c4 Bc2 cxd3 Bxd3 Bxe4 Bxe4 Nxe4 Nxd4 d5 Be3 g6 Nf3 Bf8 Qb3 Bc5 Rad1 Bxe3 Rxe3 Rc4 Qc2
    21/36	00:04	    123,961k	34,225k	+0.39	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Bd2 Rfc8 c3 c4 Bc2 cxd3 Bxd3 Bxe4 Bxe4 Nxe4 Nxd4 d5 Be3 Re8 Qg4 Qxg4 hxg4 Nd6 Nf5 Nc4 b3 Nxe3 Nxe3 Bc5 Nxd5 Rxe1+ Rxe1 Bxa3 Nc7
    22/36-	00:05	    132,174k	34,528k	+0.32	Nc3 Nd4
    22/36-	00:05	    134,573k	34,621k	+0.24	Nc3 Nd4
    22/36+	00:05	    134,973k	34,635k	+0.32	Nc3
    22/36	00:05	    140,204k	34,799k	+0.30	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Bd2 Rac8 a4 h6 axb5 axb5 Ng3 d5 e5 Nh7 Qg4 Qxg4 hxg4 c4 Nf5 Bc5 Ba2 Ra8 b4
    23/34-	00:08	    244,619k	36,494k	+0.23	Nc3 Nd4
    23/34+	00:08	    274,305k	36,824k	+0.30	Nc3
    23/34-	00:10	    353,974k	37,367k	+0.15	Nc3 Nd4
    23/34	00:10	    357,012k	37,387k	+0.20	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 d5 e5 Ne8 c3 dxc3 bxc3 g6 d4 cxd4 cxd4 a5 Ne2 Ng7 Bd2 Ne6 Bc2 Rfc8 Bd3 Bc6 Qb3 Qa7 Qb2
    24/33	00:10	    358,146k	37,393k	+0.20	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 d5 e5 Ne8 c3 dxc3 bxc3 g6 d4 cxd4 cxd4 a5 Bc2 Ng7 Bd2 Rfc8 Ne2 Ne6 Bd3 Bc6 Qb3 Qa7 Qb2 Rab8 Be3 b4 axb4
    25/33	00:11	    371,629k	37,459k	+0.17	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 d5 e5 Ne8 c3 dxc3 bxc3 g6 d4 cxd4 cxd4 a5 Ne2 Ng7 Nf4 Ne6 Bc2 Rfc8 Nxe6 Qxe6 Bd3 b4 axb4 Bxb4 Re2 Ba6 Bh6
    26/43	00:13	    449,593k	37,661k	+0.14	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 d5 e5 Ne8 c3 dxc3 bxc3 g6 Bh6 Ng7 d4 cxd4 cxd4 a5 Re3 Rfc8 Rf3 Ne6 Ne2 Bg5 Bxg5 Nxg5 Rf6 Ne6 Qd2 Qc6 Bd1
    27/41+	00:22	    828,120k	38,426k	+0.22	Nbd2
    27/41+	00:24	    899,951k	38,407k	+0.29	Nbd2
    27/41	00:29	  1,076,059k	38,443k	+0.29	Nbd2 Nd8 c3 Ne6 d4 exd4 Nxd4 c5 Nxe6 fxe6 Bc2 Rae8 Qe2 c4 a4 Qc6 Nf3 e5 Bg5 h6 Bd2 Nd7 b3 Ra8 Reb1 Nc5 axb5 axb5 bxc4 Rxa1 Rxa1
    28/34-	00:35	  1,306,496k	38,549k	+0.22	Nbd2 Nd8
    28/40-	00:41	  1,569,646k	38,625k	+0.14	Nbd2 Nd8
    28/42	00:45	  1,718,088k	38,624k	+0.14	Nbd2 Nd8 c3 Ne6 d4 exd4 Nxd4 Ng5 Bc2 Rfe8 f4 Ne6 Nxe6 Qxe6 b4 a5 Bb2 axb4 cxb4 c5 bxc5 d5 e5 Bxc5+ Kh2 Ne4 Bxe4 dxe4 Nxe4
    29/43+	01:26	  3,323,556k	38,961k	+0.22	Be3
    29/43	01:39	  3,825,421k	39,048k	+0.18	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Bd2 Rfc8 Nf4 Bd8 c4 Bb6 Qf3 Re8 a4 bxc4 Bxc4 a5 Nh5 Nxh5 Qxh5 Ba6 Bxa6 Rxa6 b3 Raa8 Qg4 Qxg4 hxg4 Rab8 Rab1 Bc7 f4 h6
    30/45	01:44	  4,050,011k	39,094k	+0.21	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Bd2 Rfc8 Nf4 Bd8 c4 Bb6 Qf3 Re8 Ba2 bxc4 Bxc4 a5 b4 cxb4 axb4 a4 b5 a3 Bb4 Ba5 Reb1 d5 Nxd5 Nxd5 Bxd5 Bxd5 exd5 Bxb4 Rxb4 Ra5
    31/44	01:53	  4,379,327k	39,104k	+0.17	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Bd2 Rfc8 Nf4 Bd8 c4 Bc7 Qf3 Re8 Nd5 Bd8 Qg3 Nxd5 cxd5 Bf6 a4 g6 f4 Bg7 f5 Rac8 Bc2 c4 axb5 Be5 Qf2 axb5 fxg6 hxg6 Ra7
    32/44+	01:59	  4,630,884k	39,120k	+0.25	Nc3
    32/45-	02:54	  6,811,856k	39,306k	+0.10	Nc3 Rfe8
    32/46	03:26	  8,069,230k	39,258k	+0.13	Nc3 Rfe8 Bd2 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Qc6 Nf5 Bf8 Qf3 d5 e5 Nd7 Bf4 a5 a4 b4 Nd6 Nxe5 Bxe5 Bxd6 Bxd5 Qxd5 Qxd5 Bxd5 Bxd6 c4 Rxe8+ Rxe8 dxc4 Bxc4 Rd1 Re2 Rxd4 Rxc2 Bc7 Bb3 Bxa5
    33/47	03:53	  9,102,977k	39,223k	+0.11	Nc3 Rfe8 Bd2 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Qc6 Nf5 Bf8 Qf3 d5 e5 Nd7 Bf4 a5 Nd6 Bxd6 exd6 Nf6 Ba2 h6 Bxh6 Qxd6 Bf4 Qd7 Bd2 a4 Qf4 Bc6 c3 dxc3 bxc3 d4 cxd4 Qxd4 Qxd4 cxd4
    34/44+	05:01	 11,748,003k	39,077k	+0.18	Nc3
    34/45	05:22	 12,537,766k	39,020k	+0.12	Nc3 Rfe8 Bd2 Nd8 a4 Ne6 Bxe6 fxe6 b4 Bc6 Ra2 Rf8 Qe2 Bb7 Qe3 c6 Ne2 h6 Rea1 Nh5 c4 Nf4 Bc3 Rad8 Rd1 Ng6 c5 Nh4 Nxh4 Bxh4 g3 Bf6 cxd6 Qxd6 Qb6 Qb8 Kg2 bxa4
    35/50	08:04	 18,738,349k	38,790k	+0.08	Nc3 Rfe8 a4 Nd4 Nxd4 exd4 Nd5 Nxd5 exd5 Bf6 Rxe8+ Rxe8 Bd2 c5 dxc6 Qxc6 Qg4 Bc8 Qh5 Be6 axb5 axb5 Bxe6 Rxe6 Qf3 Qxf3 gxf3 Re8 Bb4 Be5 Ra5 Rc8 Rxb5 g6 Rd5 Rxc2 Bxd6 Bxd6 Rxd6 Rxb2 Rxd4 Rb3 Rd7 Kg7 d4 Rxf3 Kg2 Rd3 d5 Kg8
    36/44+	08:14	 19,104,755k	38,777k	+0.15	Nc3
    36/50	09:31	 22,078,571k	38,695k	+0.16	Nc3 Rfe8 Bd2 Nd8 a4 c6 Ne2 c5 Ng3 Ne6 Ng5 Nd8 c3 h6 Nf3 Ne6 Bc2 Bf8 Nf5 Rac8 axb5 axb5 b4 c4 dxc4 Rxc4 Ra7 Qc6 Bd3 Rxe4 Bxe4 Nxe4 Qc2 Nxd2 Qxd2 Rb8 Rd1 Rd8 Ne3 g6 Qc2
    37/49+	09:51	 22,820,581k	38,665k	+0.23	Nc3
    37/49	10:28	 24,224,543k	38,627k	+0.21	Nc3 Rfe8 Bd2 Nd8 Nd5 Ne6 Nxe7+ Qxe7 Ng5 Nd8 c3 c5 Bc2 Qc7 a4 Bc6 Qe2 Qb7 a5 b4 cxb4 cxb4 Rac1 h6 Nf3 Ne6 Bb3 Nc5 Bc4 Rac8 Be3 Bb5 Nd2 Rc7 f4 exf4 Bxf4 Bxc4 Nxc4
   2017-08-21 09:23:14, Time for this analysis: 00:10:29, Rated time: 10:29
Here is the output of Texel with only 1 GB hash:

Code: Select all

Analysis of /media/danniel/sas_orig/chess/n37c.epd   
24-8-17 02:20:17 Level: 37 Ply
Analyzing engine: Texel64

 1                      
    Avoid move: 
    Best move (Texel64): Nb1-c3
    Not found in: 7:20:35
      2-	00:00	          79	20k	+0.15	Nc3 Nh5
      2-	00:00	         115	29k	-0.03	Nc3 h6
      2	00:00	         155	39k	-0.03	Nc3 h6
      3+	00:00	         291	36k	+0.25	Nc3 h6 Be3
      3	00:00	         379	47k	+0.25	Nc3 h6 Be3
      4	00:00	         740	49k	+0.15	Nc3 h6 Nd5 a5
      5+	00:00	          1k	60k	+0.32	Nc3 h6 Bd5 Nxd5 Nxd5
      5	00:00	          2k	71k	+0.13	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 exd5
      6	00:00	          3k	17k	+0.09	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 Bxd5 Bxd5 exd5
      6+	00:00	          3k	18k	+0.11	Bd5 Nxd5 exd5 Na7
      6	00:00	          3k	18k	+0.09	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 Bxd5 Bxd5 exd5
      6+	00:00	          4k	28k	+0.12	Be3 h6 Nc3 Nh5 Rf1
      6	00:00	          5k	34k	+0.09	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 Bxd5 Bxd5 exd5
      7	00:00	          6k	35k	+0.13	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 exd5 c5 dxc6 Bxc6
      7+	00:00	          6k	36k	+0.15	Be3 Na5 Ba2 h6 Nc3 Nc6 Nd5
      7+	00:00	          7k	42k	+0.31	Be3 Na5 Ba2 h6 Nc3 Nc6 Nd5
      7	00:00	          7k	44k	+0.31	Be3 Na5 Ba2 h6 Nc3 c5 Ne2 Nc6
      8-	00:00	          9k	52k	+0.13	Be3 Na5 Ba2 h6 Nc3 c5 Ne2 Nc6
      8	00:00	          9k	54k	+0.10	Be3 Na5 Ba2 h6 Nc3 c5 Nd5 Bxd5 exd5 Nh5
      9	00:00	         19k	102k	 -0.00	Be3 Na5 Ba2 c5 Nc3 Nc6 Ne2 Nd4 Nexd4 cxd4 Bd2
      9+	00:00	         22k	117k	+0.01	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 exd5 h6 c4 c5 dxc6
      9	00:00	         25k	126k	 -0.00	Be3 Na5 Ba2 c5 Nc3 Nc6 Ne2 Nd4 Nexd4 cxd4 Bd2
     10	00:00	         49k	214k	+0.07	Be3 Na5 Ba2 c5 Nc3 Nc6 Bg5 h6 Bxf6 Bxf6 Nd5 Qe6 Qd2 Nd4 Nxd4 exd4
     11+	00:00	         86k	303k	+0.22	Be3 Na5 Ba2 c5 Nc3 Nc6 b4 cxb4 Nd5 bxa3 Nb6
     11	00:00	        137k	459k	+0.24	Be3 d5 exd5 Nxd5 Bd2 Bf6 Bxd5 Qxd5 Nc3 Qd6 Ne4
     12-	00:01	        494k	1,415k	+0.09	Be3 Nd8 Ng5 c5 Rf1 h6
     12	00:01	        622k	1,660k	+0.11	Be3 Nd8 c4 c5 Nc3 Ne6 cxb5 axb5 Bxe6 fxe6 b4 cxb4 axb4 Rxa1 Qxa1 Nh5 Qa5 Bc6 d4 exd4 Nxd4 Ra8
     13	00:01	      3,161k	6,235k	+0.10	Be3 d5 exd5 Nxd5 Bxd5 Qxd5 Nc3 Qd6 a4 Nd4 Nxe5 Nxc2 Qxc2 Qxe5 d4 Qh5 axb5
     13+	00:01	      3,312k	6,468k	+0.11	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Nf5 d5
     13+	00:01	      4,388k	8,140k	+0.26	Nc3 h6 Nd5 Nxd5 Bxd5 a5 a4 b4 Rf1 Bf6 Be3 Rab8 c3 bxc3 bxc3 Ne7
     13	00:01	      4,572k	8,404k	+0.22	Nc3 h6 Nd5 Nxd5 Bxd5 a5 a4 b4 Be3 Bf6 Rf1 Rab8 b3 Nd4 Bxb7 Nxf3+
     14	00:01	      5,223k	8,720k	+0.27	Nc3 h6 Nd5 Nxd5 Bxd5 a5 a4 b4 Be3 Bf6 Rf1 Rab8 c4 Nd4 Bxd4 exd4 Qe2
     15	00:01	      6,461k	9,323k	+0.22	Nc3 h6 Nd5 a5 a4 Nxd5 Bxd5 b4 Be3 Bf6 Rf1 Rab8 c4 Nd4 Bxd4 exd4 Qe2 Bxd5 cxd5 c5
     16+	00:01	      9,601k	11,457k	+0.37	Nc3 h6 Nd5 Nxd5 Bxd5 Rab8 c3 Bf6 Nh2 Ne7 Bxb7 Rxb7 Qg4 Qxg4 Nxg4
     16	00:01	     10,242k	10,736k	+0.34	Nc3 h6 Nd5 a5 a4 Nxd5 Bxd5 b4 c3 Rab8 d4 Bf6 cxb4 axb4 Be3 Nxd4 Nxd4 Bxd5 exd5 exd4 Bxd4
     17	00:02	     25,895k	18,523k	+0.33	Nc3 h6 Be3 Na5 Ba2 Nc6 Qd2 Kh7 Nd5 a5 Nc3 b4 Nd5 Nh5 axb4 Nxb4 Nxe7 Qxe7 Bb3
     18	00:02	     50,481k	22,648k	+0.41	Nc3 h6 Be3 Na5 Ba2 Nc6 Qe2 Nd8 b4 Ne6 Nd5 Bd8 Qd2 Kh7 c4 bxc4 Bxc4 Rc8 Qc2 Nxd5 exd5
     19	00:03	     74,547k	26,056k	+0.34	Nc3 h6 Be3 Nd8 d4 Nxe4 Nd5 Ne6 dxe5 Rad8 Qe2 N4c5 Ba2 Na4 c4 bxc4 Bxc4 Bxd5 Bxd5 dxe5 Rad1 Qb5
     20	00:04	    117,946k	29,747k	+0.34	Nc3 h6 Be3 Nd8 d4 Nxe4 Nd5 Ne6 dxe5 Rad8 Qe2 N4c5 Ba2 Na4 c4 bxc4 Bxc4 Bxd5 Bxd5 dxe5 Rad1 Qb5
     21	00:06	    191,014k	31,996k	+0.27	Nc3 h6 Be3 Nd8 d4 Nxe4 Nd5 Ne6 dxe5 Rad8 c4 N4c5 Rc1 bxc4 Bxc4 Bxd5 Qxd5 c6 Qd1 d5 Bxc5 Bxc5 Bd3 a5 Qd2 Bb6 b4 axb4
     22	00:25	    877,409k	35,432k	+0.24	Nc3 Rfe8 Be3 h6 Nd5 Nxd5 exd5 Nb8 c4 Qf5 Rc1 c5 Bc2 Qc8 b4 Nd7 Qe2 cxb4 axb4 a5 cxb5 Qd8 bxa5 Qxa5 d4 Bxd5 dxe5 Nxe5 Nd2
     23	00:32	  1,115,437k	35,517k	+0.25	Nc3 Rfe8 Be3 h6 Nd5 Nxd5 exd5 Nb8 c4 c6 d4 bxc4 Ba4 Bf6 Rc1 exd4 Bxd4 Rxe1+ Qxe1 Bxd4 Nxd4
     24	00:56	  2,000,498k	36,038k	+0.37	Nc3 Rfe8 Be3 Nd8 Nd5 Nxd5 exd5 c6
     25-	01:55	  4,156,455k	36,188k	+0.22	Nc3 Rfe8 Be3 Nd8 Nd5 Nxd5 exd5 c6 dxc6 Nxc6
     25	02:03	  4,459,478k	36,180k	+0.25	Nc3 Rfe8 Be3 Nd8 Nd5 Nxd5 exd5 c6 dxc6 Nxc6 c3 h6 Nd2 Na5 Ba2 d5 Qf3 Bh4 Qg4 Qxg4 hxg4 Bc6 Nf3 Bf6 g5 hxg5 Nxg5 Rac8 Kh2 Be7 Rh1 Rf8
     26	03:34	  7,773,465k	36,395k	+0.23	Nc3 Rfe8 Be3 Nd8 d4 exd4 Bxd4 Ne6 e5 Bxf3 Bxe6 Bxd1 Bxd7 Nxd7 exd6 Bxd6 Raxd1 Rxe1+ Rxe1 f6 Kf1 Rd8 Rd1 Re8 f3 Kf7 Ne4 Bf4 Ke2 Nb6 b3 Nd5 Kd3 a5 g3 Bh6
     27+	10:49	 23,442,487k	36,137k	+0.38	Nc3 Rfe8 Ne2 Nd8 Ng3 c5 c3
     27	12:06	 26,105,659k	35,979k	+0.31	Nc3 Rfe8 Ne2 Nd8 Ng3 c5 c3 Ne6 a4 h6 Nh2 Bc6 Nf5 h5 Nxe7+ Qxe7 Bxe6 Qxe6 Bg5 g6 a5 Rab8 Bh6 b4 Nf3 bxc3 bxc3 Qb3 Qd2 c4 Ng5 Qb7
     28	21:04	 44,045,182k	34,836k	+0.28	Nc3 Rfe8 Ne2 Nd8 Ng3 c5 c3 Ne6 a4 h6 Nh2 Bc6 Nf5 h5 Nxe7+ Qxe7 Bxe6 Qxe6 Bg5 g6 a5 Rab8 Bh6 b4 Nf3 bxc3 bxc3 Qb3 Qd2 c4 Ng5 Qb7
     29	44:53	 89,619,722k	33,284k	+0.26	Nc3 Rfe8 Be3 Nd8 a4 Ne6 Bxe6 fxe6 b4 Reb8 Qb1 bxa4 Rxa4 h6 Qb3 Kh7 Qa3 Nh5 Rd1 Rf8 Qb3 Nf4 Bxf4 exf4 Ra5 Bf6 Rda1 Qe8 d4 Rb8 Kh2 Qf7
     30	1:02:16	122,557,978k	32,807k	+0.26	Nc3 Rfe8 Be3 Nd8 a4 Ne6 Bxe6 fxe6 b4 Reb8 Qb1 bxa4 Rxa4 h6 Qb2 c5 Qa2 c4 Ra5 cxd3 cxd3 Bd8 Ra3 Bc6 Rb1 Bc7 Qc2 d5 Bc5 Qf7 Rba1 d4 Ne2 Bb5 Ng3 Kh7
     31	1:33:12	181,566,299k	32,468k	+0.26	Nc3 Rfe8 Be3 Nd8 a4 Ne6 Bxe6 fxe6 b4 Reb8 Qb1 bxa4 Rxa4 h6 Qb2 Kh7 Qa3 Nh5 Ra5 Rf8 Qb3 Bd8 Raa1 Be7 Nh2 Nf4 Ra5 Rab8 Bxf4 exf4 d4 Bg5 Nf3 Be7
     32	2:24:18	278,579,508k	32,177k	+0.26	Nc3 Rfe8 Be3 Nd8 a4 Ne6 Bxe6 fxe6 b4 Reb8 Qb1 bxa4 Rxa4 h6 Qb2 Kh7 Qa3 Nh5 Ra5 Rf8 Qb3 Bd8 Raa1 Be7
     33	4:08:55	480,814,574k	32,195k	+0.26	Nc3 Rfe8 Be3 Nd8 a4 Ne6 Bxe6 fxe6 b4 Reb8 Qb1 bxa4 Rxa4 h6 Qb2 Kh7 Qa3 Nh5 Ra5 Rf8 Qb3
   2017-08-24 09:40:55, Time for this analysis: 07:20:35, Rated time: 7:20:35

0 of 1 matching moves
24-8-17 09:40:56, Total time: 07:20:39
Rated time: 7:20:35 = 26435 Seconds
And here is the output of Texel with 32 GB hash:

Code: Select all

Analysis of /media/danniel/sas_orig/chess/n37c.epd   
24-8-17 09:53:29 Level: 37 Ply
Analyzing engine: Texel64

 1                      
    Avoid move: 
    Best move (Texel64): Nb1-c3
    Not found in: 23:50:00
   Found 0 syzygy tablebases
      2-	03:35	          79	5k	+0.15	Nc3 Nh5
      2-	03:37	         115	6k	-0.03	Nc3 h6
      2	03:37	         155	7k	-0.03	Nc3 h6
      3+	03:37	         291	10k	+0.25	Nc3 h6 Be3
      3	03:37	         379	12k	+0.25	Nc3 h6 Be3
      4	03:37	         740	19k	+0.15	Nc3 h6 Nd5 a5
      5+	03:37	          1k	23k	+0.32	Nc3 h6 Bd5 Nxd5 Nxd5
      5	03:37	          2k	27k	+0.13	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 exd5
      6	03:37	          2k	34k	+0.09	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 Bxd5 Bxd5 exd5
      6+	03:37	          3k	40k	+0.19	Be3 h6 Nc3 Nh5 Bd5
      6	03:37	          4k	51k	+0.09	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 Bxd5 Bxd5 exd5
      7	03:37	          6k	62k	+0.13	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 exd5 c5 dxc6 Bxc6
      7+	03:37	          6k	64k	+0.14	Be3 d5 exd5 Nxd5 Bc1
      7	03:37	          7k	62k	+0.13	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 exd5 c5 dxc6 Bxc6
      8	03:37	         31k	218k	+0.13	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 exd5 c5 dxc6 Bxc6 Qe2
      9-	03:37	         67k	403k	-0.02	Nc3 Nd4 Nxd4 exd4 Nd5 Nxd5 exd5 h6 Qe2 Bf6
      9+	03:37	         87k	472k	+0.28	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 h6 Nf5
      9	03:37	        120k	623k	+0.18	Nc3 Nd4 Nxd4 exd4 Ne2 Rfe8 Ng3 c5 Bf4 a5 Nf5 h6
      9+	03:37	        131k	667k	+0.19	Be3 d5 exd5 Nxd5 Bd2 f6 Nc3
      9	03:37	        191k	912k	+0.18	Nc3 Nd4 Nxd4 exd4 Ne2 Rfe8 Ng3 c5 Bf4 a5 Nf5 h6
     10	03:37	        488k	2,198k	+0.27	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Bd2 Rac8 c4 h6 Qf3 Rb8
     11	03:37	        713k	3,019k	+0.32	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Bd2 h6 Nf5 d5 Nxe7+ Rxe7 e5 Nh7 Qg4 Qxg4
     12-	03:37	        805k	3,170k	+0.17	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Bd2 h6 Qf3 Rac8 c4 dxc3 Bxc3 c4 dxc4 bxc4 Ba2
     12	03:37	        983k	3,782k	+0.19	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Bd2 h6 Qf3 Rac8 c4 bxc4 dxc4 Nh7 Qf5 Nf6
     13	03:37	      1,970k	6,545k	+0.19	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Bd2 h6 Qf3 Rac8 c4 bxc4 dxc4 Nh7 Qf5 Nf6
     14	03:37	      2,659k	7,662k	+0.19	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Bd2 h6 Qf3 Rac8 c4 bxc4 dxc4 Nh7 Qf5 Qxf5 Nxf5 Bf8
     15-	03:37	      3,781k	8,555k	+0.04	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Nf4 Rfe8 Bd2 h6 c3 c4 dxc4 bxc4 Bxc4 Nxe4 Qc2 Nxd2 Qxd2 dxc3 Qxc3 Bf6
     15	03:37	      4,208k	8,712k	+0.10	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Bd2 h6 Qf3 Rac8 c4 bxc4 Bxc4 d5 exd5 Bxd5 Bxd5 Nxd5 Re5 Rcd8
     15+	03:37	      4,965k	9,871k	+0.11	Be3 Na5 Ba2 c5 Nc3 h6 b4 Nc6 bxc5 dxc5 a4 bxa4 Nd5 Nxd5 exd5
     15	03:37	      6,161k	10,605k	+0.15	Be3 Na5 Ba2 c5 Nc3 Nc6 b4 cxb4 Nd5 Bd8 Nxb4 Ba5 Rb1 h6 Qe2 Rac8 c4 Bc7 Rf1 Nxb4 axb4
     16	03:37	     10,382k	13,259k	+0.10	Be3 d5 exd5 Nxd5 Bd2 Nf4 Nxe5 Nxh3+ gxh3 Nxe5 Rxe5 Qxh3 Bd5 Bxd5 Rxd5 Bd6 Qh5
     16+	03:37	     10,837k	13,530k	+0.11	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Bd2 h6 Qf3 Rac8 c3 c4 Ba2 dxc3 Bxc3 d5 Nf5
     16+	03:37	     13,398k	16,182k	+0.26	Nc3 h6 Be3 Nh7 Bd5 Nf6 b3 Nxd5 exd5 Na5 b4
     16	03:37	     18,488k	16,962k	+0.10	Be3 d5 exd5 Nxd5 Bd2 Nf4 Nxe5 Nxh3+ gxh3 Nxe5 Rxe5 Qxh3 Bd5 Bxd5 Rxd5 Bd6 Qh5
     17	03:37	     31,020k	23,254k	+0.10	Be3 d5 exd5 Nxd5 Bxd5 Qxd5 Nc3 Qe6 Ne4 Rfd8 Qd2 f6 Qe2 Bd6 c4 bxc4 dxc4 Ne7 Nc5 Bxc5 Bxc5 Nf5 b4
     18	03:37	     38,327k	25,500k	+0.10	Be3 d5 exd5 Nxd5 Bxd5 Qxd5 Nc3 Qe6 Ne4 Rfd8 Qd2 f6 Qe2 Bd6 c4 bxc4 dxc4 Ne7 Nc5 Bxc5 Bxc5 Nf5 b4
     18+	03:37	     38,796k	25,127k	+0.11	Nc3 Nd4 Nxd4 exd4 Ne2 c5 Ng3 Rfe8 Bd2 h6 Qf3 d5 exd5 Nxd5 Qf5 Qxf5 Nxf5 Bf6 Rxe8+ Rxe8 Nd6
     18+	03:37	     40,747k	24,125k	+0.26	Nc3 h6 Be3 Nd8 Nd5 Nxd5 exd5 c6 c4 Kh7 Bb6 f5 Rc1 c5
     18	03:37	     56,953k	27,237k	+0.32	Nc3 h6 Be3 Kh7 Ba2 Kg8 Nd5 Nxd5 exd5 Nb8 c4 Qf5 Rc1 c6 Qb3 Rc8 d4 cxd5 cxb5
     19	03:38	     87,963k	30,832k	+0.32	Nc3 h6 Be3 Kh7 Ba2 Kg8 Rc1 Nd8 Nd5 Nxd5 exd5 c5 b4
     20	03:41	    193,307k	33,027k	+0.37	Nc3 Bd8 Be3 Na5 Ba2 c5 b4 cxb4 axb4 Nc6 Qb1 a5 Nxb5 axb4 c3 Ba5 cxb4 Nxb4 Bc4 Nc6 Rc1 d5 exd5 Nxd5 Bxd5 Qxd5 Rc5 Qd7 Nxe5 Nxe5
     21	03:42	    243,707k	33,312k	+0.37	Nc3 Bd8 Be3 Na5 Ba2 c5 Ne2 Nh5 b4 cxb4 axb4 Nc6 c3 a5 d4 axb4 dxe5 bxc3 Nxc3 Ba5 Ng5 Bxc3 Qxh5
     22-	03:47	    401,436k	34,069k	+0.22	Nc3 Bd8 Be3 Ne7 d4 exd4 Bxd4 Ng6 e5 dxe5 Bxe5 Qxd1 Raxd1 Bxf3 gxf3 Nh4 Re3 c5 Bg3 Nf5 Re5 Nxg3
     22	03:55	    697,443k	34,267k	+0.16	Nc3 Bd8 a4 Na5 Ba2 c5 Bg5 h6 Bxf6 Bxf6 Bd5 Nc6 axb5 axb5 Rxa8 Bxa8 Nxb5 Ne7 c4 Bxd5 exd5 Nxd5
     23	04:10	  1,204,291k	34,236k	+0.19	Nc3 Bd8 Ba2 h6 Be3 Ne7 d4 exd4 Bxd4 Nc6 Bxf6 Bxf6 e5 Nxe5 Nxe5 Bxe5 Rxe5
     24	04:32	  1,960,005k	34,278k	+0.33	Nc3 Bd8 Ba2 Ne7 Nh4 h6 Qf3 Kh7 Ne2 d5 exd5 Bxd5 Bxd5 Nfxd5 Ng3
     25	04:59	  2,903,065k	34,352k	+0.24	Nc3 Bd8 Ba2 h6 Nd5 Nd4 Nxf6+ Bxf6 Nh2 Qe7 Ng4 Bg5 Ne3 Qf6 c3 Ne6 Bxe6 fxe6 Ng4 Qg6 Bxg5 Qxg5 Qb3 Qh4 Qxe6+ Kh7 d4 exd4
     26	06:48	  6,702,601k	34,649k	+0.33	Nc3 Rfe8 Be3 Nd8 a4 Ne6 Bxe6 fxe6 b4 Rf8 Qb1 Nh5 Qb3 Nf4 Bxf4 Rxf4 axb5 axb5 Rxa8+ Bxa8 Ra1 Bc6 Ne2 Rf8 Ng3 h6 Nf5 Kh7 Nxe7 Qxe7 Ra7 Ra8 Rxa8 Bxa8
     27	07:42	  8,569,138k	34,704k	+0.34	Nc3 Rfe8 Be3 Nd8 a4 Ne6 Bxe6 fxe6 b4 Rf8 Qb1 h6 Qb3 Kh7 Red1 bxa4 Rxa4 Nh5 Rda1 Rab8 Ra5 Nf4 Bxf4 exf4
     28	14:47	 23,272,880k	34,615k	+0.23	Nc3 Rfe8 Be3 Nd8 a4 Ne6 Bxe6 fxe6 b4 h6 Qb1 Kh7 Qb3 Rf8 Ra3 bxa4 Rxa4 Nh5 Ra5 Rae8 Rea1 Nf4 Nh2 Rb8
     29	20:03	 33,516,342k	33,920k	+0.23	Nc3 Rfe8 Be3 Nd8 a4 Ne6 Bxe6 fxe6 b4 h6 Qb1 Kh7 Qb3 Rf8 Ra3 bxa4 Rxa4 Nh5 Ra5 Bd8 Raa1 Be7 Ra5
     30	31:07	 54,508,011k	32,992k	+0.24	Nc3 Rfe8 a4 Nd4 Nxd4 exd4 Ne2 c5 Ng3 h6 Qf3 d5 e5 Qe6 Bxh6 Nd7 Bxg7 Kxg7 Nf5+ Kg8 axb5 axb5 Qg4+ Qg6 Rxa8 Bxa8 e6 Qxg4 exd7
     31	1:08:49	123,826,194k	31,638k	+0.23	Nc3 Rfe8 a4
     32+	1:41:19	183,387,844k	31,275k	+0.38	Nc3 Rfe8 a4 h6 Nd5 Nxd5 Bxd5 Bf6 Be3 bxa4 c3
     32	2:41:33	294,033,311k	31,024k	+0.28	Nc3 Rfe8 a4 Nd4 Nxd4 exd4 Ne2 c5 Ng3 h6 Qf3 d5 e5 Nh7 Nf5
     33	3:48:13	414,936,698k	30,786k	+0.31	Nc3 Rfe8 a4 Na5 Ba2 c5 Ne2 Nc6 Ng3 Nd8 Bd2 Ne6 Ng5 Nd8 Ba5 Rf8 Nf5 h6 Nf3 Kh7 Bd2 Ne6 b4 c4 Be3
     34	7:47:22	844,606,634k	30,352k	+0.31	Nc3 Rfe8 a4 Na5 Ba2 c5 Bg5 b4 Bxf6 Bxf6 Nd5 Bd8 Ne3 Bc6 Nd2 Rb8 Qh5 Rf8 h4 Bf6 g3 g6 Qg4 Qxg4 Nxg4 Bg7 Nc4 Nxc4 Bxc4 Bd7 Ne3 a5 Nd5 h5 Kg2 Kh7 Rh1 f5 b3 f4
     35	19:28:23	2,172,633,338k	31,087k	+0.24	Nc3 Rfe8 Bg5 Nd4 Nxd4 exd4 Ne2 c5 Ng3 d5 e5
   2017-08-25 09:43:32, Time for this analysis: 23:50:00, Rated time: 23:50:00

0 of 1 matching moves
25-8-17 09:43:33, Total time: 23:50:03
Rated time: 23:50:00 = 85800 Seconds
One thing that is very interesting is how well Texel scales in NPS. Notice that the NPS is nearly as high as AsmFish, which tends to be much higher than Stockfish. The NPS tails off when the hash gets filled.

You will notice a strange effect on the last analysis using Texel. There is 3:35 pause before anything happens. If you watch the Arena GUI, you will see that the allocated memory is growing during this time frame. I have 128 GB RAM on the system, so 32 GB should be no problem, but for some reason allocations beyond 8GB take a while. So, for the last analysis, we should probably subtract 3.5 minutes because there is a huge start up cost.

At any rate, as we can see, increasing the RAM makes the time to ply much shorter for Texel.

Even so, it is also clear that it will take several days to get to 37 plies, compared to ten minutes for AsmFish, and a little longer for Stockfish or Komodo.

So I guess the bottom line is that NPS scaling is important, but not extremely important. Texel has incredible scaling in NPS, but not near enough to keep up with SF or Komodo.

RAM utilization is important but not extremely important. It gave a fabulous speedup to Texel. But not enough to be competitive to the top engines. The 3.5 minute lag is only for the first position in my analysis so the delay for a huge RAM setup on my machine is not significant. I often run batches for a month.

Branching factor is important, and it is extremely important. But when you cut away the fat, you have to avoid cutting away the muscle with it. And aye, there's the rub.

This may all seem a moot point because I am using long time control. But these same things are true at fast time control, they are just not quite so obvious.

Of course, reducing branching factor is easier said than done.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Top
petero2
Posts: 734
Joined: Mon Apr 19, 2010 7:07 pm
Location: Sweden
Full name: Peter Osterlund

Re: Factors in search... What is important?

Post by petero2 »

Dann Corbit wrote:You will notice a strange effect on the last analysis using Texel. There is 3:35 pause before anything happens. If you watch the Arena GUI, you will see that the allocated memory is growing during this time frame. I have 128 GB RAM on the system, so 32 GB should be no problem, but for some reason allocations beyond 8GB take a while. So, for the last analysis, we should probably subtract 3.5 minutes because there is a huge start up cost.
I guess this is caused by page allocation and page table setup in the kernel. Is this a windows computer? The development version of texel (107a29) has support for large pages on windows (2MB instead of 4KB), which should significantly reduce this overhead. However you have to configure your OS to enable large pages.

In linux there is also support for 1GB pages, which reduces the startup time to almost 0. Windows does not support 1GB pages though.

Texel initializes the hash table immediately when it is allocated, which forces all pages to be allocated at once. Stockfish behaves differently because it uses calloc to allocate the hash table, which does not force pages to be allocated. Instead they are allocated on demand when the memory is used, so the search will run very slowly for a while until all pages have been allocated. For example, on my linux computer:

Code: Select all

c24:~$ chess/sf8/stockfish | awk '$0 ~ "nps" {for (i=1; i<=NF; i++) {if ($i=="nps") nps=$(i+1); else if ($i=="time") t=$(i+1); } printf("t=%.3f nps=%d\n", t*1e-3,nps); }'
uci
setoption name hash value 32768
go infinite
t=0.016 nps=1250
t=0.021 nps=2238
t=0.056 nps=2375
t=0.219 nps=2589
t=0.484 nps=2791
t=0.905 nps=3265
t=1.488 nps=3629
t=2.051 nps=3918
...
t=4.069 nps=135732
t=4.128 nps=143382
t=4.161 nps=148033
t=4.255 nps=159795
t=4.746 nps=211810
t=4.874 nps=224692
...
t=20.181 nps=675754
...
t=32.130 nps=833362
...
t=50.754 nps=993286
...
t=80.920 nps=1120822
...
t=153.676 nps=1258095
If I insert a "ucinewgame" command before "go" to force stockfish to clear the hash table, which also causes all pages to be allocated, I instead get:

Code: Select all

c24:~$ chess/sf8/stockfish | awk '$0 ~ "nps" {for (i=1; i<=NF; i++) {if ($i=="nps") nps=$(i+1); else if ($i=="time") t=$(i+1); } printf("t=%.3f nps=%d\n", t*1e-3,nps); }'
uci
setoption name hash value 32768
ucinewgame
go infinite
t=0.003 nps=6666
...
t=0.107 nps=874457
t=0.178 nps=993050
t=0.315 nps=1082374
t=0.492 nps=1122548
t=0.597 nps=1138909
t=0.910 nps=1203464
...
t=5.236 nps=1401268
..
t=10.936 nps=1420400
...
t=30.923 nps=1427164
With texel I get full speed almost immediately after the search starts:

Code: Select all

c24:~$ ~/workspace/texel/Release/texel | awk '$0 ~ "nps" {for (i=1; i<=NF; i++) {if ($i=="nps") nps=$(i+1); else if ($i=="time") t=$(i+1); } printf("t=%.3f nps=%d\n", t*1e-3,nps); }'
uci
setoption name hash value 32768
go infinite
...
t=0.001 nps=579000
...
t=0.133 nps=1646383
...
t=1.054 nps=1578727
...
t=5.394 nps=1569102
...
t=10.731 nps=1574042
...
t=20.744 nps=1568330
Top
Dann Corbit
Posts: 12845
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Re: Factors in search... What is important?

Post by Dann Corbit »

petero2 wrote:
Dann Corbit wrote:You will notice a strange effect on the last analysis using Texel. There is 3:35 pause before anything happens. If you watch the Arena GUI, you will see that the allocated memory is growing during this time frame. I have 128 GB RAM on the system, so 32 GB should be no problem, but for some reason allocations beyond 8GB take a while. So, for the last analysis, we should probably subtract 3.5 minutes because there is a huge start up cost.
I guess this is caused by page allocation and page table setup in the kernel. Is this a windows computer? The development version of texel (107a29) has support for large pages on windows (2MB instead of 4KB), which should significantly reduce this overhead. However you have to configure your OS to enable large pages.
No. It runs Ubuntu server.
In linux there is also support for 1GB pages, which reduces the startup time to almost 0. Windows does not support 1GB pages though.
I guess that I have to do something to set up the giant pages.
Texel initializes the hash table immediately when it is allocated, which forces all pages to be allocated at once. Stockfish behaves differently because it uses calloc to allocate the hash table, which does not force pages to be allocated. Instead they are allocated on demand when the memory is used, so the search will run very slowly for a while until all pages have been allocated. For example, on my linux computer:

Code: Select all

c24:~$ chess/sf8/stockfish | awk '$0 ~ "nps" {for (i=1; i<=NF; i++) {if ($i=="nps") nps=$(i+1); else if ($i=="time") t=$(i+1); } printf("t=%.3f nps=%d\n", t*1e-3,nps); }'
uci
setoption name hash value 32768
go infinite
t=0.016 nps=1250
t=0.021 nps=2238
t=0.056 nps=2375
t=0.219 nps=2589
t=0.484 nps=2791
t=0.905 nps=3265
t=1.488 nps=3629
t=2.051 nps=3918
...
t=4.069 nps=135732
t=4.128 nps=143382
t=4.161 nps=148033
t=4.255 nps=159795
t=4.746 nps=211810
t=4.874 nps=224692
...
t=20.181 nps=675754
...
t=32.130 nps=833362
...
t=50.754 nps=993286
...
t=80.920 nps=1120822
...
t=153.676 nps=1258095
If I insert a "ucinewgame" command before "go" to force stockfish to clear the hash table, which also causes all pages to be allocated, I instead get:

Code: Select all

c24:~$ chess/sf8/stockfish | awk '$0 ~ "nps" {for (i=1; i<=NF; i++) {if ($i=="nps") nps=$(i+1); else if ($i=="time") t=$(i+1); } printf("t=%.3f nps=%d\n", t*1e-3,nps); }'
uci
setoption name hash value 32768
ucinewgame
go infinite
t=0.003 nps=6666
...
t=0.107 nps=874457
t=0.178 nps=993050
t=0.315 nps=1082374
t=0.492 nps=1122548
t=0.597 nps=1138909
t=0.910 nps=1203464
...
t=5.236 nps=1401268
..
t=10.936 nps=1420400
...
t=30.923 nps=1427164
With texel I get full speed almost immediately after the search starts:

Code: Select all

c24:~$ ~/workspace/texel/Release/texel | awk '$0 ~ "nps" {for (i=1; i<=NF; i++) {if ($i=="nps") nps=$(i+1); else if ($i=="time") t=$(i+1); } printf("t=%.3f nps=%d\n", t*1e-3,nps); }'
uci
setoption name hash value 32768
go infinite
...
t=0.001 nps=579000
...
t=0.133 nps=1646383
...
t=1.054 nps=1578727
...
t=5.394 nps=1569102
...
t=10.731 nps=1574042
...
t=20.744 nps=1568330
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Top
Dann Corbit
Posts: 12845
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Re: Factors in search... What is important?

Post by Dann Corbit »

Dann Corbit wrote:
petero2 wrote:
Dann Corbit wrote:You will notice a strange effect on the last analysis using Texel. There is 3:35 pause before anything happens. If you watch the Arena GUI, you will see that the allocated memory is growing during this time frame. I have 128 GB RAM on the system, so 32 GB should be no problem, but for some reason allocations beyond 8GB take a while. So, for the last analysis, we should probably subtract 3.5 minutes because there is a huge start up cost.
I guess this is caused by page allocation and page table setup in the kernel. Is this a windows computer? The development version of texel (107a29) has support for large pages on windows (2MB instead of 4KB), which should significantly reduce this overhead. However you have to configure your OS to enable large pages.
No. It runs Ubuntu server.
In linux there is also support for 1GB pages, which reduces the startup time to almost 0. Windows does not support 1GB pages though.
I guess that I have to do something to set up the giant pages.
Perhaps this:
https://www.howtoinstall.co/en/ubuntu/xenial/hugepages
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Top
petero2
Posts: 734
Joined: Mon Apr 19, 2010 7:07 pm
Location: Sweden
Full name: Peter Osterlund

Re: Factors in search... What is important?

Post by petero2 »

Dann Corbit wrote:I guess that I have to do something to set up the giant pages.

Perhaps this:
https://www.howtoinstall.co/en/ubuntu/xenial/hugepages
If you want to reserve 8 1GB pages on NUMA node 0, it should be enough to run the following as root:

Code: Select all

echo 8 >/sys/devices/system/node/node0/hugepages/hugepages-1048576kB/nr_hugepages
It may not actually give any performance increase though (except for startup time). I am currently running some tests to measure the effect of different combinations of page size, hash size and number of search threads.
Top

Return to “Computer Chess Club: Programming and Technical Discussions”