Lc0 benchmarks
Open command line and type there:
lc0 benchmark
Step-by-step instructions for Windows can be found here:
https://github.com/LeelaChessZero/lc0/wiki/Running-a-benchmark
* Required
Lc0 version
*
v0.23.0
Other:
Weights file
*
For standard Lc0 network files, type e.g. 42000 for network id42000. For contributor/external networks, put some descriptive names. It's recommended that you use 11248, 42850, 58467 or 58467, so that it's easy to compare different hardware in same environment.
11248
42850
58467
61325
Other:
Lc0 backend
*
For non-standard configurations (multiplexing, demux, etc), select "Other" and specify command line agruments, e.g "--backend=multiplexing --backend-opts=cudnn-fp16(gpu=0),cudnn(gpu=1)"
cudnn (Default with CUDA builds)
cudnn-fp16 (CUDA builds running with --backend=cudnn-fp16 flag)
opencl
OpenBLAS
DNNL BLAS
Other:
What is the size of the network?
*
Test60 series is 320x24, test40 is 256x20, test50 is 128x10
I don't know
320x24
256x20
128x10
Other:
Operating system
*
Windows 64-bit
Linux
MacOS
Other:
What is your hardware
*
For GPU backends specify GPU name, for example "RTX 2080 Super" or "2x RTX 2060" or "GTX 1080 (10% overclock)". For CPU backends, specify your CPU
Your answer
Number of threads
*
Default is 2, can be changed with --threads=X parameter
Your answer
If you have any other non-standard settings, specify them here
For example, --minibatch-size=512
Your answer
Nodes per second (nps)
*
From the last line of the output. For example if your output is "Benchmark final time 10.1125s calculating 2380.53 nodes per second.", type 2380.53
Your answer
Additional comments (leave empty if none)
Your answer
Submit
Never submit passwords through Google Forms.
This content is neither created nor endorsed by Google.
Report Abuse
-
Terms of Service
-
Privacy Policy
Forms