What is the expected nps for SF 16.1 on newer hardware?

[jasper.sinclair]

Hi Guenther
My 13900k gets a little more than 1.5 million (AVX2 compile)

Processor 13th Gen Intel(R) Core(TM) i9-13900K 3.00 GHz
Installed RAM 64.0 GB (63.8 GB usable)

===========================
Total time (ms) : 825
Nodes searched : 1303971
Nodes/second : 1580570

[chesskobra]

Hi Guenther
My 13900k gets a little more than 1.5 million (AVX2 compile)

Processor 13th Gen Intel(R) Core(TM) i9-13900K 3.00 GHz
Installed RAM 64.0 GB (63.8 GB usable)

===========================
Total time (ms) : 825
Nodes searched : 1303971
Nodes/second : 1580570

I am surprised because on my i7, 13 gen, 32GB RAM I am almost always getting more than 1.8M nps.

[Dann Corbit]

Here is what I get on my old, crusty threadripper 3970x using 32 gigs RAM, 30 threads, depth 20 with 5 man EGTB on standard disk drives (I am no longer using two threads per core so this is 30/32 of the machine's power).
This version was tweaked to load tablebasse files from a standard location for Ed Schroder without setting the UCI parameter (changed default).

Code: Select all

Stockfish dev-20240312-nogit by the Stockfish developers (see AUTHORS file)
info string Found 145 tablebases
bench 32767 30 20
info string Found 145 tablebases
...
Total time (ms) : 23424
Nodes searched  : 911540780
Nodes/second    : 38914821

This one is a bit newer than 16.1, but I guess it should perform very similarly.

[Dann Corbit]

Here is all 32 cores:

Code: Select all

Total time (ms) : 23567
Nodes searched  : 1017205666
Nodes/second    : 43162289

[Dann Corbit]

Same parameters, a search of the root position to 50 plies

Code: Select all

info depth 50 seldepth 66 multipv 1 score cp 13 nodes 10557248939 nps 23868762 hashfull 963 tbhits 0 time 442304 pv g1f3 g8f6 c2c4 e7e6 g2g3 d7d5 f1g2 d5c4 d1a4 c8d7 a4c4 c7c5 e1g1 d7c6 b1c3 a7a6 d2d4 b7b5 c4d3 b5b4 c3b1 b8d7 c1e3 c5d4 e3d4 f8e7 a2a3 c6e4 d3b3 b4a3 b1a3 a8b8 b3a2 e8g8 a3c4 e4d5 f1d1 b8b4 f3d2 d8c7 b2b3 d5g2 g1g2 f8a8 d2f3 b4b5 d4c3 c7b7 c4a5
bestmove g1f3 ponder g8f6

===========================
Total time (ms) : 442318
Nodes searched  : 10557407690
Nodes/second    : 23868365

[Dann Corbit]

With all 6 man files and a few 7 man files on SSD, the same search gets 171 tablebase hits

Code: Select all

info depth 49 currmove b1a3 currmovenumber 20
info depth 50 seldepth 76 multipv 1 score cp 15 nodes 17481043032 nps 22437627 hashfull 999 tbhits 171 time 779095 pv e2e4 e7e5 g1f3 b8c6 f1c4 g8f6 d2d3 f8c5 e1g1 d7d6 c2c3 e8g8 h2h3 c5b6 f1e1 h7h6 b1d2 a7a5 c4b3 c8e6 b3a4 c6e7 d3d4 e5d4 c3d4 d6d5 e4e5 f6d7 a4c2 a8c8 d2b1 f7f6 b1c3 c7c6 c3a4 b6c7 c1f4 f6e5 f4e5 e6f5 e5c7 c8c7 a4c5 f5c2 d1c2 d7c5 d4c5 e7f5 c2a4 c7e7 e1e7 d8e7 a4a5 f5h4 f3h4 e7h4 a5d2 h4c4 d2e3 d5d4 e3b3 c4b3 a2b3 f8f5 b3b4
bestmove e2e4 ponder e7e5

===========================
Total time (ms) : 779119
Nodes searched  : 17481045507
Nodes/second    : 22436939

[Vinvin]

This one is a bit newer than 16.1, but I guess it should perform very similarly.

Please pick the same exact version to perform "bench". Variation can be big.

Code: Select all

Nodes/second    :   934'076 for stockfish_24022417_x64_avx2.exe
Nodes/second    : 1'035'260 for stockfish_24030720_x64_avx2.exe
Nodes/second    :   961'763 for stockfish_23082406_x64_avx2.exe
Nodes/second    : 1'640'803 for stockfish_23070318_x64_avx2.exe

[Hai]

Finally after 15 years I could not resist anymore and migrated (most of my stuff) to a new computer, while the old one is/was still running
My new pc is a Ryzen7 5700G with 32GB Ram and it seems to have a very good price/power ratio.

Any numbers for SF 16.1, one core from startpos with deafult settings would be welcome. (if overclocked please mention about it)

Guenther

Code: Select all

maximmasiutin	 1	dcf1adb0	1.64	4096	Linux 6.5.0-21-generic	clang++ 18.1.0	3.11.6	231	dpbusdx2/38	1 minute ago
technologov	 12	amd7900a	1.55	15611	Linux 6.1.0-18-amd64	g++ 12.2.0	3.11.2	231	root_search3/188	seconds ago
teenychess	 3	664b0771	1.43	13897	Linux 6.5.0-25-generic	g++ 13.2.0	3.11.6	231	root_search3/187	seconds ago
Goatminola	 6	538a31c8	1.42	9216	Linux 6.5.0-25-generic	g++ 11.4.0	3.10.12	231	smallnetdiffscaling4/43	seconds ago
maximmasiutin	 1	9b0b70db	1.32	4096	Darwin 22.6.0	clang++ 17.0.6	3.12.2	231	dpbusdx2/40	seconds ago
robal	 	11	29b025c3	1.31	16345	Windows 10	g++ 13.2.0	3.11.5	231	corrPv/234	seconds ago
vidar808	 12	9c4d5cac	1.25	32730	Windows 10	g++ 13.2.0	3.11.8	231	dpbusdx2/35	1 minute ago
vaskoul		 6	7cbb5b7b	1.23	7978	Linux 5.15.133.1-microsoft-standard-WSL2	g++ 13.1.0	3.12.2	231	ttpvSaveTw2/2	seconds ago
rensonthemove	 3	dda55fd5	1.22	8192	Darwin 23.3.0	clang++ 15.0.0	3.9.6	231	tuned_lmr_cond5l/185	1 minute ago
sebastronomy	 31	6b074983	1.19	257697	Linux 5.14.21-150500.55.44-default	g++ 12.3.0	3.10.7	231	nn-8747f5ce199f.nnue/471	1 minute ago
mkstockfishtester 5	47419aa3	1.16	8192	Darwin 23.4.0	clang++ 15.0.0	3.9.6	231	smallnetdiffscaling4/45	1 minute ago
sebastronomy	 30	3241985e	1.15	257697	Linux 5.14.21-150500.55.44-default	g++ 12.3.0	3.10.7	231	spd4aR/75	seconds ago
SimonDemel	 8	MacBookP	1.12	32768	Darwin 23.4.0	clang++ 17.0.6	3.11.4	231	ttpvSaveTw2/0	1 minute ago

7x Linux
4x macOS
2x Windows

[Hai]

Hi Guenther
My 13900k gets a little more than 1.5 million (AVX2 compile)

Processor 13th Gen Intel(R) Core(TM) i9-13900K 3.00 GHz
Installed RAM 64.0 GB (63.8 GB usable)

===========================
Total time (ms) : 825
Nodes searched : 1303971
Nodes/second : 1580570

viewtopic.php?t=82459
Apple MacBook PRO 16-inch M1 MAX 12556848 nodes/second
Apple MacBook PRO 16-inch M3 MAX 26809.54 kn/s

16 CPU cores
40 GPU cores
128 GB RAM
8 TB SSD
…

https://www.apple.com/macbook-pro/specs/

https://www.apple.com/apple-events/

https://www.apple.com/macbook-pro/