Has anyone got LC0 18.0 to work with Arena?

[Chess_Dragon]

SUCCESS Lc0 0.28.0 with 2 GPU's - Linux, Ubuntu 21.04

Sorry for the delay. I was busy and also had 2 problems:
1. none of the instructions on the web except the link here:
https://lichess.org/forum/team-lichess- ... ecific-gpu
worked on my system.
2. I had to get a Molex-to-GPU-power adapter (my PSU just has 1 built-in).

Code: Select all

Arena 3.10 Linux: 
Backend multiplexing 
BackendOptions multi_stream=true,(backend=cuda-fp16,gpu=0),(backend=cuda-fp16,gpu=1)
  or (backend=cuda-fp16,gpu=0),(backend=cuda-fp16,gpu=1)
Threads 3  *For both GPU's at 100%: if default 2 my gpu 0 is only 94% -see screenshot*

With the default Threads 2 my GTX is so slow that with RTX+GTX its actually slower, current T60 large net.
(Not shown: For multi_stream a smaller net must be used for my GTX.
RTX+GTX about the same as RTX if Threads 3.)
The Linux terminal needs the "'s below, NOT Arena.

Code: Select all

$ ./Lc0-0.28.0 --backend=multiplexing "--backend-opts=(backend=cuda-fp16,gpu=0),(backend=cuda-fp16,gpu=1)"
       _
|   _ | |
|_ |_ |_| v0.28.0+git.unknown built Aug 28 2021
go nodes 300000
Found pb network file: ./weights_run1_610069.pb.gz
Creating backend [multiplexing]...
Creating backend [cuda-fp16]...
CUDA Runtime version: 11.2.0
Latest version of CUDA supported by the driver: 11.2.0
GPU: GeForce RTX 2060
GPU memory: 5.78766 Gb
GPU clock frequency: 1680 MHz
GPU compute capability: 7.5
Creating backend [cuda-fp16]...
CUDA Runtime version: 11.2.0
Latest version of CUDA supported by the driver: 11.2.0
GPU: GeForce GTX 1650 SUPER
GPU memory: 3.82019 Gb
GPU clock frequency: 1740 MHz
GPU compute capability: 7.5
info depth 1 seldepth 2 time 2298 nodes 7 score cp 5 nps 583 tbhits 0 pv g2g3 d7d5
... ... ...
info depth 9 seldepth 33 time 32102 nodes 261702 score cp 7 nps 8777 tbhits 0 pv d2d4 g8f6 c2c4 e7e6 g2g3 f8b4 c1d2 b4e7 d1c2 c7c6 d2f4 d7d5 g1f3 d5c4 c2c4 e8g8 f1g2
bestmove d2d4 ponder g8f6
quit

$ ./Lc0-0.28.0 -b cuda-fp16
       _
|   _ | |
|_ |_ |_| v0.28.0+git.unknown built Aug 28 2021
go nodes 300000
Found pb network file: ./weights_run1_610069.pb.gz
Creating backend [cuda-fp16]...
CUDA Runtime version: 11.2.0
Latest version of CUDA supported by the driver: 11.2.0
GPU: GeForce RTX 2060
GPU memory: 5.78766 Gb
GPU clock frequency: 1680 MHz
GPU compute capability: 7.5
info depth 1 seldepth 2 time 1622 nodes 2 score cp 6 nps 166 tbhits 0 pv c2c4 e7e5
... ... ...
info depth 9 seldepth 33 time 30402 nodes 264393 score cp 7 nps 9182 tbhits 0 pv d2d4 g8f6 c2c4 e7e6 g2g3 f8b4 c1d2 b4e7 d1c2 c7c6 d2f4 d7d5 g1f3 d5c4 c2c4 e8g8 f1g2
bestmove d2d4 ponder g8f6
quit

Terminal: $ nvidia-smi
verifies both are running, Threads=3 is needed for both to be 100%: