I used for tests Leela ver .0.25.1 what was compiled with CUDA11.1 and cudnn 7.6.4.
My system is
OS - Windows 10 x64
CPU - Ryzen 9 3950x 16 cores
NVIDIAs - 1 x RTX 2080 Ti OC, 2 x RTX 2060 OC
The used engines based on Leela version 0.25.1
Engine 1 - kDUAL (2 x RTX 2060, k refer to Kiudee params) Threads = 2 x 4000 NHz
NNCacheSize = 20000000, Backend = Mutex, BackendOptions = cudnn-fp16, Others are default
Engine 2 - kSOLO (1 x RTX 2080 Ti k refer to Kiudee params) Backend cudnn-fp16, Threads = 2 x 4000 MHz
NNCacheSize = 20000000, others are default
For TESTs I used my HalfMiniBook (50 opening position) with alternated colors and 6 men syzygy plus 5 men
Nalimov (only for GUI)
Chess GUI was Fritz 14
TC = 1 min + 2 sec / move
TEST 1
kDUAL with J92-190 against kSOLO with SV-384x30-T60-3010 = 6 : 1 (93 draw) 2x50 = 100 games
TEST 2
kDUAL with SV-384x30-T60-3010 against kSOLO with J92-190 = 1: 6 (93 draw) 2x50 = 100 games
Summarized result
J92-190 : SV-384x30-T60-3010 = 12 : 2 (186 draw) 200 games
Test match between J92-180 and SV 384x30-T60-3010
Moderators: hgm, Rebel, chrisw
-
corres
- Posts: 3657
- Joined: Wed Nov 18, 2015 11:41 am
- Location: hungary
-
corres
- Posts: 3657
- Joined: Wed Nov 18, 2015 11:41 am
- Location: hungary
Re: Test match between J92-180 and SV 384x30-T60-3010
[quote=corres post_id=866197 time=1602846045 user_id=8071]
Sorry for my foolishness, the match was between { J92-190 and SV-384x30-T60-3010 } nets.
I used for tests Leela ver .0.25.1 what was compiled with CUDA11.1 and cudnn 7.6.4.
My system is
OS - Windows 10 x64
CPU - Ryzen 9 3950x 16 cores
NVIDIAs - 1 x RTX 2080 Ti OC, 2 x RTX 2060 OC
The used engines based on Leela version 0.25.1
Engine 1 - kDUAL (2 x RTX 2060, k refer to Kiudee params) Threads = 2 x 4000 NHz
NNCacheSize = 20000000, Backend = Mutex, BackendOptions = cudnn-fp16, Others are default
Engine 2 - kSOLO (1 x RTX 2080 Ti k refer to Kiudee params) Backend cudnn-fp16, Threads = 2 x 4000 MHz
NNCacheSize = 20000000, others are default
For TESTs I used my HalfMiniBook (50 opening position) with alternated colors and 6 men syzygy plus 5 men
Nalimov (only for GUI)
Chess GUI was Fritz 14
TC = 1 min + 2 sec / move
TEST 1
kDUAL with J92-190 against kSOLO with SV-384x30-T60-3010 = 6 : 1 (93 draw) 2x50 = 100 games
TEST 2
kDUAL with SV-384x30-T60-3010 against kSOLO with J92-190 = 1: 6 (93 draw) 2x50 = 100 games
Summarized result
J92-190 : SV-384x30-T60-3010 = 12 : 2 (186 draw) 200 games
Sorry for my foolishness, the match was between { J92-190 and SV-384x30-T60-3010 } nets.
I used for tests Leela ver .0.25.1 what was compiled with CUDA11.1 and cudnn 7.6.4.
My system is
OS - Windows 10 x64
CPU - Ryzen 9 3950x 16 cores
NVIDIAs - 1 x RTX 2080 Ti OC, 2 x RTX 2060 OC
The used engines based on Leela version 0.25.1
Engine 1 - kDUAL (2 x RTX 2060, k refer to Kiudee params) Threads = 2 x 4000 NHz
NNCacheSize = 20000000, Backend = Mutex, BackendOptions = cudnn-fp16, Others are default
Engine 2 - kSOLO (1 x RTX 2080 Ti k refer to Kiudee params) Backend cudnn-fp16, Threads = 2 x 4000 MHz
NNCacheSize = 20000000, others are default
For TESTs I used my HalfMiniBook (50 opening position) with alternated colors and 6 men syzygy plus 5 men
Nalimov (only for GUI)
Chess GUI was Fritz 14
TC = 1 min + 2 sec / move
TEST 1
kDUAL with J92-190 against kSOLO with SV-384x30-T60-3010 = 6 : 1 (93 draw) 2x50 = 100 games
TEST 2
kDUAL with SV-384x30-T60-3010 against kSOLO with J92-190 = 1: 6 (93 draw) 2x50 = 100 games
Summarized result
J92-190 : SV-384x30-T60-3010 = 12 : 2 (186 draw) 200 games