Help with Scorpio 3.0.8

[Dann Corbit]

I saw in the github comments that the fix for resign was new, so I did a build from source using mingw.
The makefile failed to strip (probably because of the missing .exe) so I stripped it myself and copied it to the nn client folder

[Dann Corbit]

I am getting about 3x the throughput now.
I was getting about 20 games per minute, now about 60.
Maybe from building on my machine using gcc 10.x

[Daniel Shawul]

Yes, resign increases game generation rate by 2x times.
The other must have come from backend improvements.
Note that I get about 85 games/min using one GPU so your is still from ideal probably due to windows.
I am working on figuring out that issue as Cliff has even much more sever problems on windows than yours.

Daniel

[Daniel Shawul]

Dann, we updated the client and you need to reinstall Scorpio too, or you could just copy the windows exe
from zip file: https://github.com/dshawul/Scorpio/rele ... cts-nn.zip
and overwrite the scorpio binary you have in bin/Windows. This may be a better option than reinstalling scorpio.
There was a bug fix in "noise" parameters and which is fixed in the scorpio binary, not in the client.

Thanks

[Dann Corbit]

So far the new version has been a face-plant.
It sends zero games over and over.
I removed everything and reinstalled.
Just restarted. We will see how it goes.

[Dann Corbit]

I get cycles of this:

Code: Select all

0: **** Starting NTS session as dann ****
0:
0: <version>
0: 6
0: 6
0: </version>
0: <parameters> 800 125 120 25 0 80 30 100 0 0
0: </parameters>
0: <checksum>
0: 3646108627
0: http://scorpiozero.ddns.net/scorpiozero/nets-11/ID-0-model-3.uff
0: </checksum>
0: Skipping net download.
0: Sending 0 games to server
0: Executing job : C:\Users\dannc\chess\nn-dist-master\scripts\job-client.bat
0: Finished executing job!

[Dann Corbit]

The binary scorpio.exe exits immediately when I do this:
uci
{long list of uci commands snipped}
uciok
ucinewgame
go infinite
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
# [st = 13336ms, mt = 29250ms , hply = 0 , moves_left 10]

It does the same thing with the version I just built from source.

[Daniel Shawul]

Ok I will check what is wrong with the windows version.
Btw you have to run "scorbio.bat" not "scorpio.exe", the latter will segfault.
So do this from the bin/Windows directory

Code: Select all

scorpio.bat go quit

Cliff and Gary are able to contribute with the new binary but they are both running on linux.

[Dann Corbit]

Ok I will check what is wrong with the windows version.
Btw you have to run "scorbio.bat" not "scorpio.exe", the latter will segfault.
So do this from the bin/Windows directory

Code: Select all

scorpio.bat go quit

Cliff and Gary are able to contribute with the new binary but they are both running on linux.

Does the same thing using the batch file

Code: Select all

C:\Users\dannc\chess\nn-dist-master\Scorpio\bin\Windows>scorpio.bat
feature done=0
Number of cores 64 of 64
treeht 83886080 X 320 = 25600.0 MB
processors [16]
ht 67108864 X 16 = 1024.0 MB
eht 262144 X 8 X 16 = 64.0 MB
pht 32768 X 24 X 16 = 12.0 MB
EgbbProbe not Loaded!
loading_time = 0s
uic
Error (unknown command): uic
uci
id name Scorpio 3.0.8
id author Daniel Shawul
option name log type check default false
option name clear_hash type button
option name resign type spin default 800 min 100 max 30000
option name mt type spin default 16 min 1 max 640
option name ht type spin default 1024 min 1 max 131072
option name eht type spin default 4 min 1 max 16384
option name pht type spin default 1 min 1 max 256
option name egbb_path type string default /home/daniel/egbb/
option name egbb_files_path type string default /home/daniel/egbb/
option name nn_path type string default ../nets-scorpio/ens-net-12x128.pb
option name nn_path_e type string default ../nets-scorpio/ens-net-12x128.pb
option name nn_path_m type string default ../nets-scorpio/ens-net-12x128.pb
option name egbb_cache_size type spin default 32 min 1 max 16384
option name egbb_load_type type spin default 3 min 0 max 3
option name egbb_depth_limit type spin default 6 min 0 max 64
option name egbb_ply_limit_percent type spin default 75 min 0 max 100
option name nn_cache_size type spin default 200 min 1 max 16384
option name n_devices type spin default 1 min 1 max 128
option name device_type type combo default CPU var CPU var GPU
option name delay type spin default 0 min 0 max 1000
option name float_type type combo default HALF var FLOAT var HALF var INT8
option name nn_type type spin default 0 min 0 max 10
option name nn_type_e type spin default -1 min -1 max 10
option name nn_type_m type spin default -1 min -1 max 10
option name nn_man_e type spin default 14 min 0 max 32
option name nn_man_m type spin default 24 min 0 max 32
option name wdl_head type check default false
option name wdl_head_m type check default false
option name wdl_head_e type check default false
option name win_weight type spin default 100 min 0 max 1000
option name draw_weight type spin default 100 min 0 max 1000
option name loss_weight type spin default 100 min 0 max 1000
option name smp_type type combo default ABDADA var YBW var ABDADA var SHT
option name smp_depth type spin default 8 min 1 max 10
option name use_singular type check default true
option name use_probcut type check default false
option name singular_margin type spin default 32 min 0 max 1000
option name probcut_margin type spin default 195 min 0 max 1000
option name contempt type spin default 10 min 0 max 100
option name alphabeta_man_c type spin default 10 min 0 max 32
option name multipv type check default false
option name cpuct_base type spin default 19652 min 0 max 100000000
option name cpuct_init type spin default 125 min 0 max 1000
option name policy_temp type spin default 235 min 0 max 1000
option name cpuct_init_m type spin default 125 min 0 max 1000
option name policy_temp_m type spin default 235 min 0 max 1000
option name cpuct_init_e type spin default 125 min 0 max 1000
option name policy_temp_e type spin default 235 min 0 max 1000
option name rand_temp type spin default 100 min 0 max 1000
option name noise_alpha type spin default 30 min 0 max 100
option name noise_beta type spin default 100 min 0 max 100
option name noise_frac type spin default 25 min 0 max 100
option name noise_ply type spin default 30 min 0 max 100
option name playout_cap_rand type check default true
option name full_playouts_frac type spin default 25 min 0 max 100
option name early_stop type check default true
option name train_data_type type spin default 0 min 0 max 2
option name fpu_red type spin default 33 min -1000 max 1000
option name fpu_is_loss type check default false
option name reuse_tree type check default true
option name backup_type type spin default 6 min 0 max 7
option name ensemble type spin default 0 min 0 max 100
option name ensemble_type type spin default 0 min 0 max 1
option name min_policy_value type spin default 10 min 0 max 1000
option name frac_alphabeta type spin default 0 min 0 max 100
option name frac_freeze_tree type spin default 100 min 0 max 100
option name frac_abrollouts type spin default 20 min 0 max 100
option name frac_abprior type spin default 30 min 0 max 100
option name mcts_strategy_depth type spin default 30 min 0 max 100
option name alphabeta_depth type spin default 16 min 1 max 100
option name evaluate_depth type spin default 0 min -4 max 100
option name virtual_loss type spin default 1 min 0 max 1000
option name visit_threshold type spin default 3200 min 0 max 1000000
option name montecarlo type check default true
option name sp_resign_value type spin default 600 min 0 max 10000
option name forced_playouts type check default false
option name policy_pruning type check default false
option name treeht type spin default 25600 min 0 max 131072
uciok
ucinetgame
go infinite
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
# [st = 13336ms, mt = 29250ms , hply = 0 , moves_left 10]

[Daniel Shawul]

I do not seem to have this problem on a windows CPU machine though.
Maybe multi-gpu mode is broken somehow, but I think you updated scorpio with last version that added resign.
I haven't touched anything that affects multi-gpu mode since then. Unless you did not do a full reinstall of Scorpio (maybe you compiled
it yourself and replaced the scorpio binary ?), there shouldn't be any new stuff added since yesterday.

I have fixed one bug with windows now though. It was not passing all parameters to the self-play game generating script atleast on my machine.
CPUCT and other important parameters were taking default values of scorpio and not ones set by the server.
This could probably be the reason between windows/linux inconsistencies of the client but I am not sure.

If you don't mind could you try these steps:

a) "git pull" one commit from nn-dist
b) remove the Scorpio folder in nn-dist
c) call "client.bat" which will reinstall Scorpio and connect to server. If you don't see analysis on start position in this step
installation is wrong.

Edit:
Ah, I see egbbdll is not installed correctly in your case.

Code: Select all

EgbbProbe not Loaded!

Also the number of threads is set to just 16, it should be 196 I belive for your 2 GPUs
and installer should set it correctly.

Code: Select all

processors [16]

I think you installed the CPU version or compiled it yourself and took the scorpio.ini from github ?

Make sure you have nnprobe-windows-gpu directory under Scorpio/ folder and that it consists of nnprobe.dll and egbbdll64.dll
In any case, following the above three steps should fix your problem, and hopefully pass the right parameters as well.


Daniel