How many kn/sec makes Houdini 6 on a Ryzen7 CPU?

[Thomas Lagershausen]

Please choose the startposition.

Give it 4GB Hash.

Let it run 1 minute.

Don´t forget to tell us with how much GHZ runs your Ryzen 7 CPU.

Thx in advance

[kgburcham]

Please choose the startposition.

Give it 4GB Hash.

Let it run 1 minute.

Don´t forget to tell us with how much GHZ runs your Ryzen 7 CPU.

Thx in advance

I would also like to see the result.


here is my old 8 core: 8 x 4.3 ht off

[D] rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq -

Nalimov 6 men EGTB available - 999 MB cache
Engine: Houdini 6 Pro x64-pext (8192 MB)
by Robert Houdart
8 threads used

26/50 1:01 +0.25-- 1.d4 d5 (1.131.039.754) 18283

[Thomas Lagershausen]

Gentleman

when you buy Houdini 6 share your benchmark (kn/sec) with us - see above.

Especially when you have a Ryzen 7 CPU.

Thx in advance.

[MikeB]

2010 Mac Pro
macOS Sierra v10.12.6
12 Core (2x6 Core 3.46 Ghz)
Parallels Desktop v13
Hash set to 1024
Threads set to 12

Code: Select all

Houdini 6 Pro x64-popc
(c) 2017 Robert Houdart

info string Licensed to xxxx@xxxx.xxx
info string 1 thread used
info string 128 MB Hash

...
info time 55002 nodes 883733757 nps 16067000 tbhits 0 hashfull 998
info time 56003 nodes 900110529 nps 16072000 tbhits 0 hashfull 998
info time 57000 nodes 916724188 nps 16082000 tbhits 0 hashfull 998
info time 58003 nodes 933422421 nps 16092000 tbhits 0 hashfull 998
info time 59001 nodes 949720801 nps 16096000 tbhits 0 hashfull 998
info time 60001 nodes 966053765 nps 16100000 tbhits 0 hashfull 998
info multipv 1 depth 27 seldepth 51 score cp 16 time 60005 nodes 966092130 nps 16100000 tbhits 0 hashfull 998 pv e2e4 e7e6 d2d4 d7d5 b1d2 c7c5 g1f3 c5d4 e4d5 e6d5 f1d3 d8e7 d1e2 e7e2 d3e2 b8c6 e1g1 f8d6 f1e1 g8e7 d2b3 e8g8 b3d4 f8e8 e2d3 c8d7 d4b5 d6b8 c1d2 a7a6 b5d4
bestmove e2e4 ponder e7e6

Note: command used is "go movetime 60000" for a one minute run

[david]

hi there

can you please clarify how to obtain the benchmark ?
i understand i open a command prompt, run houdini and then use:
"go movetime 60000" for a one minute run

is that right ?
david

[MikeB]

set the number of threads ( match the real core on your system)
set the hash

Code: Select all

setoption name threads value x
setoption name hash value xxxx ( pick  1024, 2048, or 4096 depending on RAM and threads, if not sure just use 1024 to be safe)
go movetime 60000

that's it

[Houdini]

For your info, Houdini also accepts a shorter "set" command:

Code: Select all

set threads x
set hash xxxx
go movetime 60000

This works for all options that don't have spaces in their name.
Saves some typing!

[cdani]

For your info, Houdini also accepts a shorter "set" command:

Code: Select all

set threads x
set hash xxxx
go movetime 60000

This works for all options that don't have spaces in their name.
Saves some typing!

+1

Andscacs
pf = position fen
t4 = 4 threads
20 = go depth 20

I will add more.

[david]

Ryzen 1800 at 3.9 GHz
hash 8000
"info multipv 1 depth 25 seldepth 49 score cp 25 time 60004 nodes 842693852 nps 14043000 tbhits 0 hashfull 561 pv d2d4"
hash 1000
"info multipv 1 depth 25 seldepth 47 score cp 18 time 60007 nodes 911733896 nps 15193000 tbhits 0 hashfull 1000 "
Repeat 2
"info multipv 1 depth 27 seldepth 50 score cp 23 time 60002 nodes 929250348 nps 15486000 tbhits 0 hashfull 995 pv e2e4"
Repeat 3
"info multipv 1 depth 27 seldepth 56 score cp 21 time 60003 nodes 903222262 nps 15052000 tbhits 0 hashfull 998 pv d2d4"
hash of 4096
"info multipv 1 depth 27 seldepth 48 score cp 12 upperbound time 60006 nodes 868800747 nps 14478000 tbhits 0 hashfull 867"
Repeat 2
"info multipv 1 depth 27 seldepth 48 score cp 25 time 60004 nodes 886485645 nps 14773000 tbhits 0 hashfull 871 pv"
hash of 256
"info multipv 1 depth 25 seldepth 47 score cp 27 time 60005 nodes 936635052 nps 15609000 tbhits 0 hashfull 1000 pv"

So i conclude that the nps is sensitive to hash. The bigger the hash, the slower the nps. Not a fantastic effect, but at least it makes a ~10% difference in nps.

Using a Ryzen 1800x at 3.9GHz, i posted nps of 15.6-14 million nps.
david

[Houdini]

So i conclude that the nps is sensitive to hash. The bigger the hash, the slower the nps. Not a fantastic effect, but at least it makes a ~10% difference in nps.

That's correct, the bigger hash produces lower node speeds.
Using Large Pages offsets this, it's definitely recommended with hash sizes of 4 GB or higher.