ProDeo
Would you like to react to this message? Create an account in a few clicks or log in to continue.
ProDeo

Computer Chess
 
HomeHome  CalendarCalendar  Latest imagesLatest images  FAQFAQ  SearchSearch  MemberlistMemberlist  UsergroupsUsergroups  RegisterRegister  Log in  

 

 Problem with Rebel 16.2 on AMD threadripper

Go down 
+2
Chris Whittington
Dann Corbit
6 posters
Go to page : Previous  1, 2
AuthorMessage
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 7:50 am

Dann Corbit wrote:
mwyoung wrote:
Dann Corbit wrote:
mwyoung wrote:
Dann Corbit wrote:
mwyoung wrote:
Dann Corbit wrote:
mwyoung wrote:
Dann Corbit wrote:
There are two things I have found, one is a problem with Arena (Rebel 16.2 is automatically installed as a Winboard engine).
The solution to this one is simple, manually change the type to UCI.

The second problem is a bug in Rebel (I think).  My machine has 32 cores, but 64 threads.  It is AMD 3970x and normally Arena posts a warning about too many cores (e.g. with Stockfish) but lets me use all the threads that I ask for (typically between 59 and 62 threads).  However, Rebel seems to be detecting the core count and limiting to 32 threads.  Because of this, I cannot use more than half of my CPU power.

You are better and faster to use just cores anyone. And the same with stockfish.
It depends.
If I want to use my machine for anything else while I am analyzing, then I can only allocate 31/32 threads, which is 97%. But if I use threads, I can get 63/64=98%.

Sometimes I want 30 threads for engine x and 30 threads for engine y.
The cost of doing things this way is that the CPU runs hotter. But I use a 20 Amp fan to cool my machine.

You do know that percent of CPU has nothing to do with how fast Stockfish is really running.

I have many applications that definitely benefit from large thread count (some database systems, some scientific calculations, etc.)  
I use my machine for a lot more than chess.

I like a fine level of control.

I do nothing but chess.

I can say for a fact. That Stockfish and Rebel 15 have issues with over threading.

And this is easy to test.
SF seems to perform well on tcec, which uses a boatload of threads.

That is because Stockfish is the best no matter how many threads are used.

I suggest you download the PGN files from TCEC as I did, and see how fast stockfish is running on their system. And compare the time to depth times with your system.

You will be shocked!
I may do that, it should be a fairly simple experiment.

My machine is atypical. I have 128gb ram, a gigantic raid disk farm, and 8TB of PCie 4.0 SSD.  I also have two Nvidia 2080 super gpus. This is very useful if I want (for example) 60 threads of SMP (perhaps with two different engines) and lc0 using my gpus and two or three threads.

I also do things that hurt performance, since  I want the right answers. For instance, a huge Tablebase system probably subtracts Elo, but once in a while it causes better analysis.

I guess if I used only bitbase, there would be better performance, but I am far more concerned about correctness. That's why I am horrified by engines that skip some under-promotions.

The following SQL query:
select acd, sum(acs) / count(acs) as aveacs from dbo.Tcec where Engine like 'Stockfish%' and acd >= 27 group by acd order by acd

Produced this data {first column is depth in plies for Stockfish, second column is average time to ply in seconds}:
27,12
28,18
29,23
30,29
31,37
32,45
33,54
34,63
35,68
36,75
37,78
38,79
39,78
40,80
41,76
42,74
43,74
44,71
45,70
46,70
47,72
48,71
49,71
50,68
51,70
52,67
53,67
54,67
55,62
56,62
57,58
58,55
59,59
60,55
61,54
62,56
63,51
64,53
65,52
66,51
67,54
68,53
69,53
70,49
71,49
72,47
73,48
74,50
75,50
76,50
77,51
78,55
79,45
80,50
81,48
82,50
83,52
84,50
85,54
86,52
87,46
88,48
89,51
90,49
91,50
92,49
93,43
94,44
95,52
96,47
97,47
98,51
99,48
100,43
101,50
102,46
103,48
104,50
105,40
106,45
107,38
108,39
109,47
110,45
111,41
112,50
113,47
114,39
115,45
116,55
117,33
118,42
119,38
120,32
121,39
122,35
123,42
124,40
125,34
126,35
127,8
128,40
129,34
130,45
131,37
132,46
133,30
134,42
135,39
136,29
137,44
138,28
139,75
140,55
141,46
142,56
143,48
144,38
145,36
146,44
147,36
148,52
149,38
150,55
151,44
152,44
153,50
154,37
155,31
156,37
157,48
158,27
159,59
160,32
161,45
162,39
163,30
164,39
165,64
166,48
167,27
168,37
169,43
170,39
171,20
172,40
173,25
174,49
175,21
176,29
177,24
178,30
179,32
180,28
181,28
182,45
183,34
184,28
185,26
186,16
187,46
188,29
189,32
190,23
191,101
192,57
193,24
194,24
195,81
196,33
197,20
198,18
199,67
200,40
201,12
202,42
203,25
204,39
205,18
206,25
207,30
208,57
209,14
210,19
211,10
212,38
213,25
214,14
215,45
216,72
217,15
218,20
219,31
220,23
221,25
222,27
223,123
224,44
225,24
226,26
227,113
228,79
229,23
230,20
231,95
232,27
233,27
234,12
235,44
236,23
237,25
238,98
239,61
240,30
241,28
242,26
243,31
244,51
Back to top Go down
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 7:55 am

This query:
select Engine, sum(acd) / count(acd) as aveacd from dbo.Tcec where Engine like 'Stockfish%' group by Engine order by Engine

shows the average depth during games reached by each engine with a name like Stockfish* :
Stockfish 38
Stockfish 010918 40
Stockfish 021113 38
Stockfish 021115 40
Stockfish 030916 35
Stockfish 041017 39
Stockfish 051117 41
Stockfish 060314 40
Stockfish 070915 43
Stockfish 071015L 38
Stockfish 080414 37
Stockfish 090319 37
Stockfish 091114 40
Stockfish 1 13
Stockfish 1.8 31
Stockfish 1.9 31
Stockfish 1.9.1 31
Stockfish 100218 41
Stockfish 110616 36
Stockfish 120413 37
Stockfish 128GiB_202008260719_nn-82215d0fd0df 53
Stockfish 14.1 44
Stockfish 14.1_20211101 56
Stockfish 14_202107131735 52
Stockfish 14_202108051641 61
Stockfish 14_202108220915 47
Stockfish 141214 37
Stockfish 15.1 54
Stockfish 15.1TB6 17
Stockfish 151013 37
Stockfish 160518 40
Stockfish 160716 37
Stockfish 160913 35
Stockfish 160914 39
Stockfish 170514 37
Stockfish 180214 42
Stockfish 180614 41
Stockfish 180614 Sufi 12 28
Stockfish 18102108 42
Stockfish 18102108 Sufi 13 27
Stockfish 181224 42
Stockfish 18122721 41
Stockfish 19011415 38
Stockfish 190203 48
Stockfish 190203 Sufi 14 25
Stockfish 19022119 36
Stockfish 19040612 42
Stockfish 19042711 41
Stockfish 19050918 51
Stockfish 19061419 35
Stockfish 190826 48
Stockfish 190826_dll_fix 41
Stockfish 19092522 46
Stockfish 19100908 45
Stockfish 191107 49
Stockfish 191113 38
Stockfish 191114 40
Stockfish 191118 43
Stockfish 191121 43
Stockfish 2 24
Stockfish 2.0.1 30
Stockfish 2.01 30
Stockfish 2.11 30
Stockfish 2.31 31
Stockfish 2019112601 36
Stockfish 20191201 35
Stockfish 20191203 50
Stockfish 20191209 39
Stockfish 2019121008 39
Stockfish 202003040717 63
Stockfish 202003092246 56
Stockfish 20200407DC 55
Stockfish 202004181536 51
Stockfish 202005232210 55
Stockfish 202006170741 58
Stockfish 20200701 42
Stockfish 202007032109 66
Stockfish 2020071122 49
Stockfish 20200816_nn-82215d0fd0df 34
Stockfish 202008260719_nn-82215d0fd0df 52
Stockfish 202009282242_nn-baeb9ef2d183 61
Stockfish 202010181354_nn-eba324f53044 49
Stockfish 2020102823_nn-2eb2e0707c2b 55
Stockfish 20201225 65
Stockfish 20210113 67
Stockfish 202102202249 61
Stockfish 20210310 61
Stockfish 202104151245 54
Stockfish 2021061923 50
Stockfish 210213 32
Stockfish 210219[4cores] 33
Stockfish 210516 37
Stockfish 231014 44
Stockfish 250313 37
Stockfish 250413 36
Stockfish 260318 46
Stockfish 260318 Sufi 11 31
Stockfish 270918 43
Stockfish 300414 39
Stockfish 300815 38
Stockfish 301114 37
Stockfish 4 37
Stockfish 6 Sufi 6 27
Stockfish 8 40
Stockfish 8 Sufi 9 28
Stockfish copy 20210113 68
Stockfish dev-20230114-3d2381d7 50
Stockfish dev15_20211007 54
Stockfish dev15_20211015 60
Stockfish dev15_202112071815 62
Stockfish dev15_20220302 54
Stockfish dev15_20220401 54
Stockfish dev16_202204220817 45
Stockfish dev16_20220504 59
Stockfish dev16_2022051413 51
Stockfish dev16_20220515 45
Stockfish dev16_20220521 55
Stockfish dev16_202205291914 45
Stockfish dev16_202208061357 56
Stockfish dev16_202208061357_copy 59
Stockfish dev16_20220817 40
Stockfish dev16_202209170930 50
Stockfish dev16_202209270754 35
Stockfish dev16_20221027 55
Stockfish dev16_202211232145 62
Stockfish dev16_2022121920 54
Stockfish dev16_202301021914 49
Stockfish dev16_20230114 48
Stockfish Latest 70
Stockfish_11_CCRL 64-bit_4CPU 41
Stockfish_13_CCRL 64-bit_4CPU 47
Stockfish_15_100k 18
Stockfish_15_100M 37
Stockfish_15_10k 12
Stockfish_15_10M 29
Stockfish_15_1G 50
Stockfish_15_1k 7
Stockfish_15_1M 23
Stockfish_15_300k 21
Stockfish_15_300M 42
Stockfish_15_300M scaled_to_450Mnodes 37
Stockfish_15_30k 15
Stockfish_15_30M 32
Stockfish_15_3k 9
Stockfish_15_3M 26
Stockfish_15_CCRL 64-bit_4CPU 40
Stockfish_19121008c0 32
Stockfish_21032717_16th 32
Stockfish_21032717_1thd 26
Stockfish_21032717_4ths 30
Stockfish0.1pct 20191225 10
Stockfish0.3pct 20191225 15
Stockfish0_1pct 20191225 11
Stockfish100kn 202104151245 22
Stockfish10kn 202104151245 17
Stockfish10Mn 202104151245 45
Stockfish10pct 19050918 35
Stockfish10pct 202003092246 37
Stockfish1kn 202104151245 11
Stockfish1Mn 202104151245 36
Stockfish1pct 20200112 22
Stockfish1pct 2020011312 21
Stockfish1thread 202003092246 22
Stockfish25pct dev16_202204220817 37
Stockfish3pct 202001231912 26
StockfishClassical 202007311012 50
Stockfishd1 202104151245 1
StockfishDepth1 202007172028 1
StockfishDepth2 202007172028 2
StockfishDepth3 202007172028 3
StockfishDepth4 202007172028 4
StockfishNNUE 20200704-StockFiNN-0.2 54
StockfishNoobbook 20190925 42
StockfishSimul 18
Back to top Go down
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 7:58 am

This query shows the average depth reached for games as a function of the exact engine name:
select Engine, sum(acd) / count(acd) as aveacd from dbo.Tcec where Engine like 'Stockfish%' group by Engine order by sum(acd) / count(acd) 

StockfishDepth1 202007172028 1
Stockfishd1 202104151245 1
StockfishDepth2 202007172028 2
StockfishDepth3 202007172028 3
StockfishDepth4 202007172028 4
Stockfish_15_1k 7
Stockfish_15_3k 9
Stockfish0.1pct 20191225 10
Stockfish1kn 202104151245 11
Stockfish0_1pct 20191225 11
Stockfish_15_10k 12
Stockfish 1 13
Stockfish_15_30k 15
Stockfish0.3pct 20191225 15
Stockfish10kn 202104151245 17
Stockfish 15.1TB6 17
Stockfish_15_100k 18
StockfishSimul 18
Stockfish1pct 2020011312 21
Stockfish_15_300k 21
Stockfish1thread 202003092246 22
Stockfish100kn 202104151245 22
Stockfish1pct 20200112 22
Stockfish_15_1M 23
Stockfish 2 24
Stockfish 190203 Sufi 14 25
Stockfish_15_3M 26
Stockfish3pct 202001231912 26
Stockfish_21032717_1thd 26
Stockfish 18102108 Sufi 13 27
Stockfish 6 Sufi 6 27
Stockfish 180614 Sufi 12 28
Stockfish 8 Sufi 9 28
Stockfish_15_10M 29
Stockfish_21032717_4ths 30
Stockfish 2.11 30
Stockfish 2.01 30
Stockfish 2.0.1 30
Stockfish 1.9 31
Stockfish 1.9.1 31
Stockfish 260318 Sufi 11 31
Stockfish 2.31 31
Stockfish 1.8 31
Stockfish 210213 32
Stockfish_15_30M 32
Stockfish_19121008c0 32
Stockfish_21032717_16th 32
Stockfish 210219[4cores] 33
Stockfish 20200816_nn-82215d0fd0df 34
Stockfish 20191201 35
Stockfish 160913 35
Stockfish 030916 35
Stockfish10pct 19050918 35
Stockfish dev16_202209270754 35
Stockfish 19061419 35
Stockfish 19022119 36
Stockfish 2019112601 36
Stockfish 110616 36
Stockfish1Mn 202104151245 36
Stockfish 250413 36
Stockfish 250313 37
Stockfish 210516 37
Stockfish 080414 37
Stockfish 120413 37
Stockfish 4 37
Stockfish 151013 37
Stockfish_15_300M scaled_to_450Mnodes 37
Stockfish 301114 37
Stockfish_15_100M 37
Stockfish 090319 37
Stockfish10pct 202003092246 37
Stockfish 160716 37
Stockfish 170514 37
Stockfish 141214 37
Stockfish25pct dev16_202204220817 37
Stockfish 19011415 38
Stockfish 021113 38
Stockfish 071015L 38
Stockfish 300815 38
Stockfish 191113 38
Stockfish 38
Stockfish 160914 39
Stockfish 20191209 39
Stockfish 300414 39
Stockfish 2019121008 39
Stockfish 041017 39
Stockfish 160518 40
Stockfish dev16_20220817 40
Stockfish 191114 40
Stockfish_15_CCRL 64-bit_4CPU 40
Stockfish 060314 40
Stockfish 010918 40
Stockfish 8 40
Stockfish 021115 40
Stockfish 091114 40
Stockfish 051117 41
Stockfish 190826_dll_fix 41
Stockfish 100218 41
Stockfish 18122721 41
Stockfish 19042711 41
Stockfish_11_CCRL 64-bit_4CPU 41
Stockfish 180614 41
Stockfish 18102108 42
Stockfish 20200701 42
Stockfish_15_300M 42
Stockfish 181224 42
StockfishNoobbook 20190925 42
Stockfish 180214 42
Stockfish 19040612 42
Stockfish 191118 43
Stockfish 270918 43
Stockfish 070915 43
Stockfish 191121 43
Stockfish 14.1 44
Stockfish 231014 44
Stockfish dev16_202204220817 45
Stockfish dev16_20220515 45
Stockfish dev16_202205291914 45
Stockfish 19100908 45
Stockfish10Mn 202104151245 45
Stockfish 260318 46
Stockfish 19092522 46
Stockfish 14_202108220915 47
Stockfish_13_CCRL 64-bit_4CPU 47
Stockfish dev16_20230114 48
Stockfish 190203 48
Stockfish 190826 48
Stockfish dev16_202301021914 49
Stockfish 202010181354_nn-eba324f53044 49
Stockfish 2020071122 49
Stockfish 191107 49
Stockfish dev-20230114-3d2381d7 50
Stockfish 20191203 50
Stockfish_15_1G 50
StockfishClassical 202007311012 50
Stockfish 2021061923 50
Stockfish dev16_202209170930 50
Stockfish dev16_2022051413 51
Stockfish 202004181536 51
Stockfish 19050918 51
Stockfish 14_202107131735 52
Stockfish 202008260719_nn-82215d0fd0df 52
Stockfish 128GiB_202008260719_nn-82215d0fd0df 53
Stockfish dev15_20220401 54
Stockfish 15.1 54
StockfishNNUE 20200704-StockFiNN-0.2 54
Stockfish dev15_20211007 54
Stockfish dev15_20220302 54
Stockfish 202104151245 54
Stockfish dev16_2022121920 54
Stockfish 20200407DC 55
Stockfish 2020102823_nn-2eb2e0707c2b 55
Stockfish 202005232210 55
Stockfish dev16_20220521 55
Stockfish dev16_20221027 55
Stockfish 202003092246 56
Stockfish 14.1_20211101 56
Stockfish dev16_202208061357 56
Stockfish 202006170741 58
Stockfish dev16_202208061357_copy 59
Stockfish dev16_20220504 59
Stockfish dev15_20211015 60
Stockfish 202009282242_nn-baeb9ef2d183 61
Stockfish 202102202249 61
Stockfish 14_202108051641 61
Stockfish 20210310 61
Stockfish dev15_202112071815 62
Stockfish dev16_202211232145 62
Stockfish 202003040717 63
Stockfish 20201225 65
Stockfish 202007032109 66
Stockfish 20210113 67
Stockfish copy 20210113 68
Stockfish Latest 70
Back to top Go down
Admin
Admin
Admin


Posts : 2496
Join date : 2020-11-17
Location : Netherlands

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 8:30 am

Dann Corbit wrote:
AMD Ryzen Threadripper 3970X (62 threads used by Stockfish, Test, 32 threads used by Rebel)

Meaning, the tryout patch did not work?
Back to top Go down
http://rebel13.nl/
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 8:37 am

Admin wrote:
Dann Corbit wrote:
AMD Ryzen Threadripper 3970X (62 threads used by Stockfish, Test, 32 threads used by Rebel)

Meaning, the tryout patch did not work?
It works fine.  I tested it with many different thread counts.
Stockfish and Test are using 62 threads.
Rebel is using 32.
Currently:
   Engine             Score         St        Lc        Te        Lc        Re    S-B
1: Stockfish-x64-avx2 23.0/35 ········· ======1== ===1===== ====11=1= =1111=11   364.00
2: Lc0-T78            17.5/35 ======0== ········· ========  ========= =====1===  300.00
3: Test-62t           16.5/34 ===0===== ========  ········· ========= ========   290.00
4: Lc0-T60            16.0/36 ====00=0= ========= ========= ········· ====0====  278.00
5: Rebel-16.2         14.0/34 =0000=00  =====0=== ========  ====1==== ·········  239.00


87 of 600 games played
Name of the tournament: Tdog
Site/ Country: DESKTOP-426Q3JG, United States
Level: Blitz 5/2
Hardware: AMD Ryzen Threadripper 3970X 32-Core Processor with 63.9 GB Memory
Operating system: Windows 10 Home Home Edition (Build 9200) 64 bit
PGN-File: C:\Program Files (x86)\Arena\Tournaments\Tdog.pgn
Website: 
E-Mail Address: 
Back to top Go down
Admin
Admin
Admin


Posts : 2496
Join date : 2020-11-17
Location : Netherlands

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 8:52 am

Meaning, Rebel with 62 threads is now fine?
Back to top Go down
http://rebel13.nl/
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 9:31 am

Admin wrote:
Meaning, Rebel with 62 threads is now fine?
Yes, the program test is running with the magic parameter and 62 threads.
It gets about 25% more nodes and {initial results show} better time to ply on average.
I am running a tournament because "the proof of the pudding is in the eating."
Therefore, the only meaningful question is, "Is it stronger?"
Interestingly, 63 threads was too many.  That is why I am using 62.
I guess that all the services on my machine use more than one core, so there was some kind of collision with 63 threads.
I made two attempts at analyzing a position with 63 threads.  The first time it was horribly slow.  The second time it crashed.  However, 62 seems fine and dandy on my machine.

Admin likes this post

Back to top Go down
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 9:38 am

Dann Corbit wrote:
Admin wrote:
Meaning, Rebel with 62 threads is now fine?
Yes, the program test is running with the magic parameter and 62 threads.
It gets about 25% more nodes and {initial results show} better time to ply on average.
I am running a tournament because "the proof of the pudding is in the eating."
Therefore, the only meaningful question is, "Is it stronger?"
Interestingly, 63 threads was too many.  That is why I am using 62.
I guess that all the services on my machine use more than one core, so there was some kind of collision with 63 threads.
I made two attempts at analyzing a position with 63 threads.  The first time it was horribly slow.  The second time it crashed.  However, 62 seems fine and dandy on my machine.
To be absolutely clear, the program test-62t is the program test that you sent me, with the magic cookie supplied on the command line.
It is using 62 threads.
The program Rebel-16.2 is the standard Rebel 16.2 that is downloaded from your site.
The two versions of LC0 are the latest version with the recommended T60 and T78 networks.
Stockfish is the bleeding edge stockfish.
Back to top Go down
Admin
Admin
Admin


Posts : 2496
Join date : 2020-11-17
Location : Netherlands

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 10:31 am

That's pretty cool. I will discuss it with Chris if we make it an UCI option, default=off, plus a warning on the website.
Back to top Go down
http://rebel13.nl/
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 10:34 am

Admin wrote:
That's pretty cool. I will discuss it with Chris if we make it an UCI option, default=off, plus a warning on the website.
I guess after a few hundred games we'll have a pretty good idea what improvement (if any) there is.

There is probably a way to detect if AMD systems are set to two threads per core or one.

So far it's a fun experiment. Thanks for humoring an old whinging twit.
Laughing

Admin likes this post

Back to top Go down
mwyoung

mwyoung


Posts : 880
Join date : 2020-11-25
Location : USA

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 11:31 am

Dann Corbit wrote:
Admin wrote:
That's pretty cool. I will discuss it with Chris if we make it an UCI option, default=off, plus a warning on the website.
I guess after a few hundred games we'll have a pretty good idea what improvement (if any) there is.

There is probably a way to detect if AMD systems are set to two threads per core or one.

So far it's a fun experiment. Thanks for humorist an old whinging twit.
Laughing

It will be interesting to see if you gain anything using SMT. I never have other then gaining more nodes per second using SMT. But hurting the overall search. So I never use SMT with chess engines.

Unless the chess engine is using a MCTS.
Back to top Go down
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 11:37 am

mwyoung wrote:
Dann Corbit wrote:
Admin wrote:
That's pretty cool. I will discuss it with Chris if we make it an UCI option, default=off, plus a warning on the website.
I guess after a few hundred games we'll have a pretty good idea what improvement (if any) there is.

There is probably a way to detect if AMD systems are set to two threads per core or one.

So far it's a fun experiment. Thanks for humorist an old whinging twit.
Laughing

It will be interesting to see if you gain anything using SMT. I never have other then gaining more nodes per second using SMT. But hurting the overall search. So I never use SMT with chess engines.

Unless the chess engine is using a MCTS.
I guess if I measured carefully I would see greater energy usage and maybe a 1 or 2% loss for using smt to do 60 threads instead of 30 cores for chess. If I do smt with two instances at 30 threads for twice as long I expect to see no loss except power use.

As I have explained before, I like the flexibility and I have some apps that like threads.

Purely for chess, it probably isn't worth it.
Back to top Go down
Eelco

Eelco


Posts : 205
Join date : 2021-10-08

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 5:04 pm

Dann Corbit wrote:

Purely for chess, it probably isn't worth it.

I haven't any measurements, but my feeling is, this is an understatement these days. We can't compare with Rebel 15, because only the net from Chris and the tuning by Ed will have any similarity to the ones in Rebel 16.2. So the tests Mark made will have to be redone. But talking about gut feelings, I think it is a waste of electricity these days to use more than one thread Razz

Just to get a first impression of SMP I have made a double thread Rebel and it is like a completely different program in the single testposition I used, and where the goal was to show a losing score for White. The one thread version won, in my completely subjective opinion. It is very subjective of course. I did a long time analysis from the testposition from Mark's match posted on Talkchess a while ago. The double thread version does show a NPS that seems, very roughly, double the one thread version, but for the rest all similarity seems to disappear Smile This will be mostly random, but still. Notice the extreme differences in achieved depths between the two, but at the same time how much longer it takes to get a negative score with the double version. My recommendation would still be, in the days of NNUE, get a little more memory and wait a little longer but just use all those threads for something else. One thread is enough, moreover it may just be better for analysis.

This is the long analysis with one thread. Rebel finds Nc4 earlier but that is not really important at all. Look at the depths also Rebel one thread finds a score of -1.13 after exactly 5 minutes. To reach the exact same score, the two thread version needs 73 minutes, 8 seconds. This is all statistically completely irrelevant of course, but I prefer the one thread version of Rebel 16.2. Let alone a 30 core version..

One thread analysis to depth 53 Smile

Code:

rb4r1/pp2q2k/7p/1P3p2/P2P1p1n/N1P1pP2/2Q1R1PP/2N2RK1 w - -

Engine: Rebel-16.2 (512 MB)
gemaakt door Ed Schroder and Chris Whittington (C) 2

24/49  0:01   0.00     31.Db3 Ld6 32.Pc2 Pxg2 33.Txg2 Txg2+
                       34.Kxg2 Tg8+ 35.Kh1 e2 36.Tg1 Txg1+
                       37.Kxg1 e1T+ 38.Pxe1 Dxe1+ 39.Kg2 Dd2+
                       40.Kh3 Kg7 41.De6 Dxc1 42.Dd7+ Kg6
                       43.Dxd6+ Kh5 44.Kg2 (2.060.535)

25/48  0:02   0.00     31.Db3 Ld6 32.Pc2 Pxg2 33.Txg2 Txg2+
                       34.Kxg2 Tg8+ 35.Kh1 e2 36.Tg1 Txg1+
                       37.Kxg1 e1T+ 38.Pxe1 Dxe1+ 39.Kg2 Dd2+
                       40.Kh3 Kg7 41.De6 Dxc1 42.Dd7+ Kg6
                       43.Dxd6+ Kh5 44.Kg2 (2.461.984)

26/48  0:02   0.00     31.Db3 Ld6 32.Pc2 Pxg2 33.Txg2 Txg2+
                       34.Kxg2 Tg8+ 35.Kh1 e2 36.Tg1 Txg1+
                       37.Kxg1 e1T+ 38.Pxe1 Dxe1+ 39.Kg2 Dd2+
                       40.Kh3 Kg7 41.De6 Dxc1 42.Dd7+ Kg6
                       43.Dxd6+ Kh5 44.Kg2 (2.906.069)

27/46  0:02   0.00     31.Db3 Ld6 32.Pc2 Pxg2 33.Txg2 Txg2+
                       34.Kxg2 Tg8+ 35.Kh1 e2 36.Tg1 Txg1+
                       37.Kxg1 e1T+ 38.Pxe1 Dxe1+ 39.Kg2 Dd2+
                       40.Kh3 Kg7 41.De6 Dxc1 42.Dd7+ Kg6
                       43.Dxd6+ Kh5 44.Kg2 (3.506.608)

28/48  0:03   0.00     31.Db3 Ld6 32.Pc2 Pxg2 33.Txg2 Txg2+
                       34.Kxg2 Tg8+ 35.Kh1 e2 36.Tg1 Txg1+
                       37.Kxg1 e1T+ 38.Pxe1 Dxe1+ 39.Kg2 Dd2+
                       40.Kh3 Kg7 41.De6 Dxc1 42.Dd7+ Kg6
                       43.Dxd6+ Kh5 44.Kg2 (4.582.858)

29/47  0:21   -0.66++  31.Db3 (27.805.562)

29/49  0:35   -0.69    31.Db3 Ld6 32.Pc2 Tg5 33.Kh1 Tag8
                       34.Pe1 Th5 35.Dd5 Pg6 36.h3 Pf8
                       37.Kh2 Lb8 38.Pcd3 Kh8 39.c4 Dh4
                       40.Th1 b6 41.Df7 Pg6 42.c5 Dg3+
                       43.Kg1 (45.999.767)

30/49  0:52   -0.79++  31.Db3 (66.681.507)

30/43  0:53   -0.57++  31.Db3 (68.534.066)

30/51  0:56   -0.22++  31.Db3 (71.629.675)

30/53  0:59   0.00     31.Db3 Ld6 32.Pc2 Tg5 33.Kh1 Tag8
                       34.Pe1 Th5 35.Pcd3 Tg7 36.b6 axb6
                       37.Dxb6 Pg6 38.Pf2 Pe5 39.Ph3 Pc4
                       40.Db5 Pd2 41.Tg1 De6 42.Pd3 Txh3
                       43.gxh3 Txg1+ 44.Kxg1 (75.662.961)

31/42  1:01   0.00++   31.Db3 Lc7 32.Kh1 Tg7 (78.348.492)

31/51  1:03   -0.22++  31.Db3 Tg6 (81.042.677)

31/52  1:13   -0.24++  31.Db3 Lc7 32.Kh1 Tg7 (93.656.345)

31/50  1:21   -0.79    31.Db3 Ld6 32.Pc2 Tg5 33.Pe1 Tag8
                       34.Pcd3 Th5 35.Dd5 Pg6 36.h3 Lb8
                       37.c4 Kh8 38.Kh1 Td8 39.Dc5 Dxc5
                       40.dxc5 Lc7 41.Kh2 La5 42.c6 bxc6
                       43.bxc6 Lxe1 44.Pxe1 (103.385.626)

32/56  2:17   -0.89++  31.Da2 (174.848.752)

32/47  2:33   -0.91++  31.Da2 Lc7 32.Pc2 (195.782.256)

32/51  2:41   -0.85    31.Da2 Ld6 32.Pc4 Lc7 33.Pd3 Tg7
                       34.a5 Tag8 35.Pe1 Dd7 36.b6 axb6
                       37.axb6 Lb8 38.Pa3 Da4 39.Kh1 Tg6
                       40.Tb2 Tc8 41.c4 Tcg8 42.Tg1 Dd1
                       43.Pb5 Pxf3 44.Tb1 (206.501.635)

33/52  2:53   -0.75++  31.Da2 Lc7 32.Pc2 (223.153.586)

33/46  2:57   -0.80++  31.Da2 Lc7 32.Pd3 (227.501.054)

33/49  3:18   -0.57++  31.Da2 (256.334.584)

33/53  5:00   -1.13    31.Da2 Lc7 32.Pc2 Tg5 33.Pd3 Tag8
                       34.Pce1 Th5 35.Pc5 Pg6 36.h3 Ld6
                       37.Dd5 b6 38.Pcd3 Ph4 39.c4 Td8
                       40.Kh2 Lb8 41.Dc6 Txd4 42.Pb4 Td6
                       43.Da8 Td8 44.Pd5 (387.965.701)

34/63  6:23   -1.03++  31.Db2 (496.507.679)

34/52  9:07   -1.42++  31.Da2 (703.695.071)

34/55  9:14   -0.91++  31.Da2 (712.588.847)

34/54  9:28   -1.18    31.Da2 Lc7 32.Pc2 Tg5 33.Pd3 Tag8
                       34.Pce1 Th5 35.Kh1 Dg5 36.Pb4 Tg7
                       37.Pd5 Pg6 38.b6 Lxb6 39.Pd3 Kh8
                       40.Tb1 Dg3 41.h3 Ld8 42.c4 Txh3+
                       43.gxh3 Dxf3+ 44.Kh2 (731.218.889)

35/63  19:10  -1.59++  31.Pc4 (1.467.212.130)

35/53  20:26  -1.07++  31.Pc4 (1.573.321.590)

35/64  22:53  -1.35    31.Pc4 Lc7 32.a5 Df7 33.Pb2 Tg7
                       34.Pbd3 Tag8 35.Pe1 Lxa5 36.Pb3 Lc7
                       37.c4 b6 38.Kh1 Ld6 39.c5 Le7 40.Pc1 Dd5
                       41.Pb3 bxc5 42.dxc5 Tb8 43.Dd3 Td8
                       44.Dc3 (1.768.876.188)

36/53  23:18  -1.25++  31.Pc4 Tg5 32.b6 (1.805.143.848)

36/62  23:46  -1.15++  31.Pc4 (1.845.192.386)

36/66  28:47  -1.45++  31.Pc4 Lc7 (2.238.261.173)

36/65  32:06  -0.94++  31.Pc4 (2.510.824.617)

36/64  32:30  -1.36    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8D (2.540.614.124)

37/59  33:25  -1.47++  31.Pc4 Lc7 (2.609.292.164)

37/51  33:28  -1.56    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8D (2.614.164.715)

38/56  33:59  -1.46++  31.Pc4 Lc7 (2.654.811.493)

38/56  34:13  -1.36++  31.Pc4 (2.673.341.109)

38/58  34:39  -1.21++  31.Pc4 (2.709.501.029)

38/60  36:06  -1.32++  31.Pc4 Lc7 (2.822.459.507)

38/59  38:03  -1.71    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8T (2.965.591.284)

39/57  38:34  -1.61++  31.Pc4 Tg7 32.Kh1 (3.005.656.485)

39/54  38:50  -1.51++  31.Pc4 (3.025.949.485)

39/52  39:06  -1.36++  31.Pc4 (3.047.508.849)

39/61  39:58  -1.28    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8T (3.116.243.082)

40/56  43:06  -1.51++  31.Pc4 Lc7 32.Pb2 (3.334.905.503)

40/69  48:14  -1.17++  31.Pc4 (3.719.328.801)

40/67  58:53  -1.62    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8T (4.509.992.232)

41/59  59:21  -1.52++  31.Pc4 Lc7 (4.545.390.171)

41/64  60:14  -1.42++  31.Pc4 (4.610.932.975)

41/64  60:53  -1.27++  31.Pc4 (4.661.740.497)

41/63  61:56  -1.31    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8T (4.737.165.709)

42/59  67:46  -1.41++  31.Pc4 Lc7 32.b6 (5.169.892.687)

42/59  68:05  -1.41    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8T (5.194.658.621)

43/61  69:58  -1.41++  31.Pc4 Lc7 32.Pb2 (5.336.170.546)

43/63  71:22  -1.45    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8T (5.439.911.845)

44/64  93:54  -1.55++  31.Pc4 Lc7 32.b6 (7.092.132.556)

44/65  97:38  -1.33++  31.Pc4 (7.391.479.856)

44/61  98:58  -1.65    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8T (7.488.927.382)

45/73  117:43 -1.65++  31.Pc4 Lc7 32.Pb2 (8.881.443.893)

45/72  122:03 -1.75    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8D (9.207.113.213)

46/73  126:57 -1.65++  31.Pc4 Tg7 32.Kh1 (9.587.829.722)

46/72  137:02 -1.55++  31.Pc4 (10.372.981.423)

46/69  141:15 -1.65    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8D (10.680.737.151)

47/74  145:41 -1.55++  31.Pc4 Lc7 32.Pb2 (11.019.967.000)

47/65  169:24 -1.60++  31.Pc4 Lc7 32.b6 (12.758.279.815)

47/62  249:25 -1.64++  31.Pc4 Tg7 32.Kh1 (18.598.032.841)

47/75  253:34 -1.67    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8D (18.899.114.908)

48/73  265:46 -1.57++  31.Pc4 Lc7 32.Kh1 (19.806.473.680)

48/78  288:33 -1.68    31.Pc4 Tg5 32.a5 De8 33.Kh1 Dxb5
                       34.Db3 Dxb3 35.Pxb3 Lc7 36.d5 Tag8
                       37.Pd4 Txg2 38.Txg2 Txg2 39.d6 Ld8
                       40.Tg1 Tf2 41.Pe5 Lf6 42.d7 Td2
                       43.Pf7 e2 44.d8D (21.452.768.267)

49/77  338:37 -1.68++  31.Pc4 Tg7 32.Kh1 (25.098.449.786)

49/90  698:00 -1.91++  31.Pc4 Lc7 32.Pd3 (50.769.305.563)

49/83  714:49 -1.97    31.Pc4 Tg5 32.a5 Lc7 33.b6 axb6
                       34.axb6 Ld6 35.Pd3 Tag8 36.Pe1 Lb8
                       37.Db3 Pg6 38.Da3 Dd8 39.Da8 Tg7
                       40.Kh1 Th5 41.h3 Ph4 42.Pe5 Thg5
                       43.P5d3 Tg3 44.Da2 (51.981.021.905)

50/105 1113:53-1.97++  31.Da2 (78.386.769.271)

50/85  1174:50-1.82++  31.Da2 (82.735.505.874)

50/81  1267:59-1.59++  31.Da2 (89.413.762.889)

50/116 3549:25-2.90    31.Da2 Lc7 32.b6 axb6 33.Dc4 Tac8
                       34.Pc2 Ld6 35.Dd3 Dc7 36.Pa2 Df7
                       37.Pe1 Tc4 38.Dc2 Dd5 39.Kh1 Ta8
                       40.Dd3 Tcxa4 41.Pc1 T4a5 42.Tg1 b5
                       43.Tb2 h5 44.Pc2 (244.989.899.532)

50/116 3571:09-2.20++  31.Pc4 (246.565.672.634)

50/115 3660:42-1.77    31.Pc4 Tg5 32.a5 Lc7 33.b6 axb6
                       34.axb6 Ld6 35.Pd3 Tag8 36.Pe1 Lb8
                       37.Da2 T8g7 38.Da8 Df7 39.Da4 Th5
                       40.Pe5 De7 41.P1d3 Pg6 42.Pxg6 Txg6
                       43.Pb4 Kh8 44.Db5 (252.999.220.581)

51/85  3778:40-1.77++  31.Pc4 Lc7 32.Kh1 (261.054.381.017)

51/87  3866:49-1.79++  31.Pc4 Lc7 32.Pd3 (267.105.320.990)

51/88  4008:14-1.93    31.Pc4 Tg5 32.a5 Lc7 33.b6 axb6
                       34.axb6 Ld6 35.Pd3 Tag8 36.Pe1 Lb8
                       37.Da2 T8g7 38.Da8 Df7 39.Da4 Dg6
                       40.Da2 Df6 41.Da8 Dc6 42.Da2 Th5
                       43.Pe5 Lxe5 44.dxe5 (276.737.518.837)

52/88  4156:13-2.04++  31.Pc4 Lc7 32.Kh1 (286.454.705.391)

52/81  4375:50-2.05++  31.Pc4 Lc7 32.Pd3 (300.013.860.057)

52/92  4486:20-1.97    31.Pc4 Tg5 32.a5 Lc7 33.b6 axb6
                       34.axb6 Ld6 35.Pd3 Tag8 36.Pe1 Lb8
                       37.Da2 T8g7 38.Da8 Df7 39.Da4 Dg6
                       40.Da2 Df6 41.Da8 Dc6 42.Da2 Th5
                       43.Pe5 Lxe5 44.dxe5 (306.507.701.607)

53/100 5343:30-2.20++  31.Pc4 Lc7 32.Kh1 (361.190.989.633)

beste zet: Pa3-c4  tijd: 5574:05.875 min nodes: 361.190.989.633


And this is still running, it is not very long underway, the double thread version of Rebel, it has just found 31. Nc4 and a score of -1.30 Cool :

Code:

rb4r1/pp2q2k/7p/1P3p2/P2P1p1n/N1P1pP2/2Q1R1PP/2N2RK1 w - -

Engine: Rebel-16.2 2 thread (512 MB)
gemaakt door Ed Schroder and Chris Whittington (C) 2

21/44  0:01   -0.22++  31.Db3 Lc7 32.Pc2 (3.953.312)

21/41  0:01   -0.25    31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tg6 35.Pe1 Dc4 36.Dxc4 Txc4
                       37.Tc2 Txa4 38.c4 b6 39.Ped3 Tg8
                       40.c5 Lc7 41.d5 bxc5 42.Pxc5 Ta3
                       43.Pe6 Le5 44.Tg1 (4.212.508)

22/33  0:01   -0.48++  31.Db3 Tg6 (5.020.288)

22/39  0:02   -0.14++  31.Db3 (5.158.746)

22/42  0:02   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.c4 Pxg2
                       34.Txg2 Txg2+ 35.Kxg2 Tg8+ 36.Kh1 e2
                       37.Pxe2 Dxe2 38.Tg1 Txg1+ 39.Kxg1 Le7
                       40.c5 Kg7 41.Pb4 Lf6 42.Dc3 Dd1+
                       43.Kg2 De2+ (5.488.923)

23/45  0:02   -0.22++  31.Db3 Lc7 32.Pc2 (5.863.097)

23/55  0:02   -0.53    31.Db3 Ld6 32.Pc2 Tac8 33.c4 Pxg2
                       34.Txg2 Txg2+ 35.Kxg2 Tg8+ 36.Kh1 e2
                       37.Pxe2 Dxe2 38.Tg1 Txg1+ 39.Kxg1 Le7
                       40.c5 Kh8 41.Pb4 Lh4 42.Pd3 Dxf3
                       43.De6 Dxd3 44.Dxh6+ (6.095.135)

24/46  0:02   -0.53++  31.Db3 (7.018.473)

24/33  0:02   -0.38++  31.Db3 (7.118.924)

24/44  0:03   -0.15++  31.Db3 (8.221.469)

24/46  0:03   -0.51    31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tg6 35.Pe1 Dc4 36.Dxc4 Txc4
                       37.Tc2 Tg8 38.a5 Ta4 39.c4 Txa5
                       40.Ped3 b6 41.Tg1 Tc8 42.Pe2 Txb5
                       43.Tc3 Ta5 44.c5 (9.093.001)

25/45  0:03   -0.41++  31.Db3 Lc7 32.Pc2 (9.406.996)

25/44  0:03   -0.31++  31.Db3 (9.568.779)

25/39  0:03   -0.18    31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tg6 35.Pe1 Dc4 36.Dxc4 Txc4
                       37.Tc2 Tg8 38.a5 Ta4 39.c4 Txa5
                       40.Ped3 b6 41.Tg1 Tc8 42.Pe2 Txb5
                       43.Tc3 Ta5 44.c5 (9.991.517)

26/46  0:04   -0.08++  31.Db3 Lc7 32.Pc2 (10.970.042)

26/34  0:05   -0.13++  31.Db3 Lc7 32.Pc2 (13.356.923)

26/43  0:08   -0.47++  31.Db3 (22.354.463)

26/47  0:08   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tgd8 35.De6 Kg7 36.g3 fxg3
                       37.Pxe3 Df7 38.Dxf7+ Kxf7 39.hxg3 Lxg3
                       40.Pd1 Tc4 41.Pd3 Txa4 42.f4 Ta3
                       43.Tg1 Tg8 44.Pe5+ (23.147.534)

27/43  0:09   0.00++   31.Db3 Lc7 32.Pc2 (24.408.886)

27/48  0:09   -0.02++  31.Db3 Lc7 32.Pc2 (24.761.938)

27/40  0:09   -0.05++  31.Db3 Lc7 32.Pc2 (25.463.081)

27/45  0:10   -0.26    31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tg6 35.Pe1 Dc4 36.Dxc4 Txc4
                       37.Tc2 Tg8 38.a5 Ta4 39.c4 Txa5
                       40.Ped3 b6 41.Tg1 Tc8 42.Pe2 Txb5
                       43.Tc3 Ta5 44.c5 (25.961.954)

28/50  0:10   -0.16++  31.Db3 Lc7 32.Pc2 (26.556.446)

28/47  0:10   -0.06++  31.Db3 (26.771.398)

28/48  0:11   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tgd8 35.De6 Tg8 (29.124.818)

29/49  0:12   0.00++   31.Db3 Lc7 32.Pc2 (33.498.431)

29/44  0:13   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tgd8 35.De6 Tg8 (33.825.283)

30/47  0:13   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tgd8 35.De6 Tg8 (36.105.190)

31/47  0:16   -0.10++  31.Db3 Lc7 32.Kh1 b6 (41.540.646)

31/45  0:17   -0.11++  31.Db3 Lc7 32.Kh1 Tg7 (44.035.939)

31/61  0:21   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 De7 (54.936.147)

32/55  0:23   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 b6 35.Tg1 De7 36.Db3 Dc7 (59.457.104)

33/54  0:25   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 b6 35.Tg1 De7 36.Db3 Dc7 (66.398.350)

34/63  0:36   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 b6 35.Tg1 De7 36.Db3 Dc7 (93.191.907)

35/57  0:47   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tcd8 35.Db3 (120.816.099)

36/46  0:52   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tcd8 35.Db3 (132.732.268)

37/59  1:06   0.00     31.Db3 Ld6 32.Pc2 Tac8 33.Kh1 Dc7
                       34.Dd5 Tcd8 35.Db3 (168.951.987)

38/57  3:29   -0.66++  31.Db3 Lc7 32.Kh1 Pxg2 (519.998.629)

38/81  8:01   -1.10    31.Db3 Ld6 32.Pc2 Tg5 33.Pd3 Tag8
                       34.Pce1 Th5 35.Kh1 b6 36.c4 Pg6
                       37.Pf2 Df6 38.Ph3 Dxd4 39.Pd3 Tg7
                       40.Td1 Df6 41.Pe1 Lb8 42.Pg1 Pe5
                       43.Dc2 Ld6 44.Ph3 (1.201.203.022)

39/65  16:32  -1.00++  31.Da2 (2.512.107.096)

39/55  17:02  -0.90++  31.Da2 (2.589.946.119)

39/56  18:28  -0.75++  31.Da2 (2.814.529.416)

39/62  19:18  -0.64    31.Da2 Ld6 32.Pc4 Lc7 33.Pd3 Tg7
                       34.Kh1 Tag8 35.Pe1 Df7 36.b6 axb6
                       37.Pa3 Dxa2 38.Txa2 Lb8 39.Pb5 Tc8
                       40.Pd3 h5 41.Tc1 Tc4 42.Pb4 Kh6
                       43.Tb2 Tg6 44.Te1 (2.946.346.122)

40/54  20:08  -0.86++  31.Da2 Lc7 32.Pc2 (3.073.678.510)

40/78  37:21  -1.06    31.Db3 Ld6 32.Pc2 Tg5 33.Kh1 Tag8
                       34.Pe1 Th5 35.Pcd3 b6 36.a5 Tg7
                       37.Pf2 Pxg2 38.Pxg2 Dg5 39.Pd3 Dg3
                       40.Pgxf4 Lxf4 41.Pxf4 Dxf4 42.axb6 Thg5
                       43.Db1 axb6 44.Dc1 (5.677.791.084)

41/72  40:48  -0.96++  31.Db3 Lc7 32.Kh1 Pxg2 (6.205.145.742)

41/75  43:47  -0.86++  31.Db3 (6.650.176.213)

41/72  71:23  -1.45++  31.Da2 (10.668.514.830)

41/67  73:08  -1.13    31.Da2 Lc7 32.Pc2 Tg5 33.Kh1 Tag8
                       34.Pe1 Th5 35.Pcd3 Dg5 36.Pb4 Pg6
                       37.Pd5 Kh8 38.h3 Ld6 39.c4 Pf8 40.c5 Lb8
                       41.Pd3 Pe6 42.Tfe1 Dh4 43.b6 axb6
                       44.cxb6 (10.931.709.572)

42/70  91:39  -1.53++  31.Pc4 (13.722.108.206)

42/80  109:50 -1.30    31.Pc4 Lc7 32.a5 Df7 33.Da2 Tg7
                       34.Pd3 Tag8 35.Pe1 Lb8 36.Pa3 Dxa2
                       37.Txa2 Tc8 38.Pb1 Ld6 39.Kh1 h5
                       40.a6 b6 41.Pd3 Tcg8 42.Pe1 Pg6
                       43.Pd3 Tc8 44.Tc1 (16.376.553.268)

pgn of Marks game:

[pgn]
[Event "*"]
[Site "*"]
[Date "2023.02.05"]
[Round "1"]
[White "Berserk 10 "]
[Black "Lc0 v0.29.0 Net 809106"]
[ECO "C50"]
[Result "0-1"]

1. e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. d3 Bc5 5. c3 {C50: Giuoco
Pianissimo} d6 6. O-O Bb6 {End of opening} 7. Bg5 {+0.7/31
74025 1823532748} h6 {+0.0/15 26129 310595 289/420/291}
8. Bh4 {+0.6/29 8485 218547403} g5 {+0.0/16 9372 346289
295/393/312} 9. Bg3 {+0.8/26 17786 469390992} Ne7 {+0.0/16
13238 470032 295/394/311} 10. a4 {+0.6/27 8735 222648283}
c6 {+0.0/15 35673 684616 273/424/303} 11. Na3 {+0.5/28
31730 822949917} Ng6 {+0.1/12 13035 169615 258/446/296}
12. Ba2 {+0.4/26 12929 339225432} Bc7 {+0.1/14 22478 413694
251/443/306} 13. Nc2 {+0.5/27 8315 215647910} O-O {+0.1/12
8581 230882 242/478/280} 14. Ne3 {+0.6/27 20568 547262161}
Kg7 {+0.1/11 12553 230858 243/475/282} 15. b4 {+0.4/29
36932 979540794} d5 {+0.0/14 16916 148275 271/465/264}
16. Qc2 {+0.2/27 7443 195043780} Nh5 {-0.0/14 5120 117616
270/473/257} 17. Rae1 {+0.7/24 6005 153421056} Nhf4
{-0.0/16 7604 145199 250/519/231} 18. exd5 {+0.5/28 13453
341469455} cxd5 {-0.0/23 14605 123456 262/484/254} 19. d4
{+0.3/31 19228 498301628} e4 {-0.0/19 13235 123440
281/459/260} 20. Nd2 {+0.2/31 14634 386857625} f5 {+0.0/18
10457 222855 269/468/263} 21. f3 {+0.4/29 5310 140310916}
Bd6 {-0.0/17 23531 425518 240/531/229} 22. b5 {+0.3/27 8680
223628742} Bb8 {+0.0/18 13923 257916 279/442/279} 23. Bxf4
{+0.3/31 48376 1262519267} gxf4 {-0.1/19 7817 136290
298/444/258} 24. Nxd5 {+0.2/28 5499 152137078} e3 {-0.1/18
6179 201000 298/448/254} 25. Nb1 {+0.2/26 4991 137892627}
Be6 {-0.1/17 7386 274781 299/453/248} 26. Nb4 {+0.3/26 4724
133069300} Bxa2 {-0.1/16 8135 362176 295/459/246} 27. Nxa2
{+0.2/28 11498 320958173} Kh7 {-0.1/13 33868 976357
333/412/255} 28. Re2 {+0.6/25 6935 192619088} Nh4 {-0.1/13
8523 359790 333/410/257} 29. Na3 {-0.0/38 76887 2151058031}
Rg8 {-0.1/14 12256 281952 352/406/242} 30. Nc1 {+0.0/35
9049 248959177} Qe7 {-0.1/15 13617 345341 354/401/245}
31. Nc4 {+0.0/33 4245 114182519} Bc7 {-0.5/16 14515 176019
453/401/146} 32. Nb2 {+0.2/28 8407 232385949} Rg5 {-0.8/14
13980 227383 605/257/138} 33. Nbd3 {-3.4/31 41186
1081677614} Rag8 {-0.8/14 6575 300931 615/252/133} 34. Ne1
{-3.6/25 3708 94565754} Rh5 {-0.9/12 2138 313125
619/251/130} 35. Ned3 {-3.9/31 21265 552889342} Ng6
{-3.0/14 21133 237537 861/100/39} 36. Nf2 {-3.9/26 2414
59773336} Qg5 {-4.1/14 37069 520365 898/70/32} 37. Ncd3
{-6.1/24 23647 465239355} Ne7 {-4.3/15 225 124880
902/68/30} 38. Ng4 {-6.8/23 3687 93971264} Qh4 {-4.8/15
6091 157385 914/58/28} 39. h3 {-6.8/20 1560 40057345} Kh8
{-8.2/12 14743 166538 951/33/16} 40. Qd1 {-7.7/26 13238
304068980} fxg4 {-9.7/15 14467 121739 961/23/16} 41. fxg4
{-7.7/20 1016 25123041} Rxg4 {-11.2/16 6570 161502
966/21/13} 42. Rf3 {-12.1/22 10038 234754243} Rg3 {-20.5/10
21911 207637 983/11/6} 43. Rxg3 {-13.1/22 3730 88670961}
Qxg3 {-24.1/9 9892 104597 986/8/6} 44. Qe1 {-13.8/21 3739
91413617} Qg5 {-31.4/8 22189 276948 990/6/4} 45. Nf2
{-12.9/22 8161 181929643} Nf5 {-39.1/7 13968 146739
992/6/2} 46. Ng4 {-8.6/17 635 12958821} Rh4 {-29.3/7 10187
119532 988/9/3} 47. Qf1 {-12.1/22 9068 200832717} Rxg4
{-25.2/8 12494 124162 986/9/5} 48. hxg4 {-17.2/21 6004
138923394} Ng3 {-29.0/9 4243 59740 988/8/4} 49. Rxe3
{-18.3/18 1320 30698500} Nxf1 {-57.9/8 7030 103921 996/3/1}
50. Re8+ {-18.7/19 3277 78383299} Kg7 {-50.3/7 13910 203656
995/3/2} 51. Kxf1 {-22.9/21 9311 201675272} Ba5 {-47.7/6
14335 225544 995/3/2} 52. c4 {-17.1/18 4339 101080657} Qg6
{-62.0/6 8488 86913 996/3/1} 53. Re2 {-24.4/18 6570
127729968} Bc3 {-68.1/6 8347 97454 997/2/1} 54. Re7+
{-27.7/18 5402 137312612} Kf6 {-63.5/6 7786 79087 997/2/1}
55. Re2 {M-7/36 3981 107664709} Qd3 {-49.3/6 8221 93384
995/3/2} 56. Kf2 {-33.0/15 1981 56561659} Kg5 {-40.0/6 7283
83698 990/10/0} 57. Re8 {M-7/36 3506 99961830} Kh4 {-43.2/6
7440 78841 991/9/0} 58. g3+ {-35.9/16 1372 42396462} Kh3
{M+6/6 1049 10022 1000/0/0} 59. gxf4 {M-5/69 11969
180625054} Bxd4+ {M+5/1 10 1920 1000/0/0} 60. Ke1 {M-4/73
6747 2707920} Kg3 {M+4/1 10 582 1000/0/0} 61. Re2 {M-3/78
5435 1696453} Bc3+ {M+1/1 10 2 1000/0/0} 0-1
[/pgn]
Back to top Go down
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 6:54 pm

   Engine             Score             St            Lc            Te            Lc            Re    S-B
1: Stockfish-x64-avx2 32.5/51 ············· ======1====== ===1=======1= ====11=1==1== =1111=11===1   768.25
2: Lc0-T78            27.0/51 ======0====== ············· =========1==  ============= =====1===1=1=  671.75
3: Test-62t           24.0/51 ===0=======0= =========0==  ············· ============= =============  616.50
4: Lc0-T60            23.5/52 ====00=0==0== ============= ============= ············· ====0========  603.75
5: Rebel-16.2         21.0/51 =0000=00===0  =====0===0=0= ============= ====1======== ·············  536.75


128 of 600 games played

Eelco likes this post

Back to top Go down
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 7:00 pm

Some impressions:
I have never seen a program like Rebel 16.2 which disdains his own pawn shelter and gets away with it.  Before castling, he will shred his pawn shelter on both sides and stay in the middle.
After same side castling, he will push pawns from his pawn shelter as though he was planning a pawn storm using his own shelter pawns.  But I did not see him get punished for this.

I would describe the play as "daring".

Eelco likes this post

Back to top Go down
Admin
Admin
Admin


Posts : 2496
Join date : 2020-11-17
Location : Netherlands

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 7:09 pm

Dann Corbit wrote:
Some impressions:
I have never seen a program like Rebel 16.2 which disdains his own pawn shelter and gets away with it.  Before castling, he will shred his pawn shelter on both sides and stay in the middle.
After same side castling, he will push pawns from his pawn shelter as though he was planning a pawn storm using his own shelter pawns.  But I did not see him get punished for this.

I would describe the play as "daring".

We do a lot with what is called king safety.

Mclane and Dann Corbit like this post

Back to top Go down
http://rebel13.nl/
Chris Whittington




Posts : 1254
Join date : 2020-11-17
Location : France

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 7:47 pm

Admin wrote:
That's pretty cool. I will discuss it with Chris if we make it an UCI option, default=off, plus a warning on the website.

RelaxThreadLimit = true/false

done



Back to top Go down
Chris Whittington




Posts : 1254
Join date : 2020-11-17
Location : France

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptySat Feb 25, 2023 7:49 pm

Dann Corbit wrote:
Some impressions:
I have never seen a program like Rebel 16.2 which disdains his own pawn shelter and gets away with it.  Before castling, he will shred his pawn shelter on both sides and stay in the middle.
After same side castling, he will push pawns from his pawn shelter as though he was planning a pawn storm using his own shelter pawns.  But I did not see him get punished for this.

I would describe the play as "daring".

King Shelter?! Pfffftt. WTF is King Shelter?!

Back to top Go down
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptyMon Feb 27, 2023 2:20 am

After Windows update, I have a problem on my machine now.  My machine frequently reboots when under load.  So I need to get to the bottom of it.  it did some other bad things, like completely disable my SQL Server database where I store all of my data (I did get that one fixed, though).
Here is the current standings, where the tournament left off after the last reboot:


   Engine             Score                    St                   Lc                   Lc                   Te                   Re    S-B
1: Stockfish-x64-avx2 45.5/76 ···················· ======1============  ====11=1==1========  ===1=======1=======  =1111=11===1=1=====   1643.2
2: Lc0-T78            40.0/76 ======0============  ···················· ===================  =========1=======1=  =====1===1=1=======   1497.2
3: Lc0-T60            36.5/77 ====00=0==0========  ===================  ···················· ===================  ====0==========1====  1393.2
4: Test-62t           36.0/76 ===0=======0=======  =========0=======0=  ===================  ···················· ===================   1387.0
5: Rebel-16.2         33.0/77 =0000=00===0=0=====  =====0===0=0=======  ====1==========0==== ===================  ····················  1277.2


191 of 600 games played
Name of the tournament: Tdog
Site/ Country: DESKTOP-426Q3JG, United States
Level: Blitz 5/2
Hardware: AMD Ryzen Threadripper 3970X 32-Core Processor with 63.9 GB Memory
Operating system: Windows 10 Home Home Edition (Build 9200) 64 bit
PGN-File: C:\Program Files (x86)\Arena\Tournaments\Tdog.pgn
Website: 
Back to top Go down
Dann Corbit




Posts : 164
Join date : 2020-11-26

Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 EmptyThu Mar 09, 2023 12:36 pm

I never did find out what  was causing the crashes, but eventually, my machine settled down.  I hate mysteries like that, where it is fixed by magic.  It makes you wonder when the other shoe is going to drop-.
We are a little over half way now.

Code:

 [font=monospace]  Engine             Score                                      St                                    Lc                                    Lc                                    Te                                    Re    S-B
1: Stockfish-x64-avx2 87.0/146 ····································· ======1===================1========== ====11=1==1==========1============1=  ===1=======1=======1===1===1=1=1=1=1= =1111=11===1=1=====1===========1=1==   5965.0
2: Lc0-T78            78.0/145 ======0===================0========== ····································· ====================================  =========1=======1====1===========1=  =====1===1=1===========1=1=====1=111   5496.5
3: Lc0-T60            70.5/145 ====00=0==0==========0============0=  ====================================  ····································· ================================1===  ====0==========1=============1=======  5098.7
4: Test-62t           65.5/145 ===0=======0=======0===0===0=0=0=0=0= =========0=======0====0===========0=  ================================0===  ····································· ====================================   4815.7
5: Rebel-16.2         62.0/145 =0000=00===0=0=====0===========0=0==  =====0===0=0===========0=0=====0=000  ====1==========0=============0======= ====================================  ·····································  4588.5
[/font]
[size]


363 of 600 games played
Name of the tournament: Tdog
Site/ Country: DESKTOP-426Q3JG, United States
Level: Blitz 5/2
Hardware: AMD Ryzen Threadripper 3970X 32-Core Processor with 63.9 GB Memory
Operating system: Windows 10 Home Home Edition (Build 9200) 64 bit
PGN-File: C:\Program Files (x86)\Arena\Tournaments\Tdog.pgn
[/size]

Eelco likes this post

Back to top Go down
Sponsored content





Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty
PostSubject: Re: Problem with Rebel 16.2 on AMD threadripper   Problem with Rebel 16.2 on AMD threadripper - Page 2 Empty

Back to top Go down
 
Problem with Rebel 16.2 on AMD threadripper
Back to top 
Page 2 of 2Go to page : Previous  1, 2
 Similar topics
-
» The Stockfish ELO problem
» Small Somu 1.7a problem ...
» New test set "Problem of the Week"
» Rebel in my FACE -Tournaments
» Rebel 14.2

Permissions in this forum:You cannot reply to topics in this forum
ProDeo :: Computer Chess-
Jump to: