Skip to content

rl-tools/l2f-benchmark

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

git clone https://github.com/rl-tools/l2f-benchmark
git submodule update --init -- external/rl_tools
nvcc benchmark.cu -Iexternal/rl_tools/include -use_fast_math --optimize 3 -std=c++17 && ./a.out

Note that the optimal values for N_BLOCKS, N_THREADS and N_ITERATIONS vary depending on your GPU. These values were tuned for the Nvidia T2000 but I got about 20 billion steps per second or ~6.4 years of simulated time per second on an RTX 4090 (mobile). I believe with better tuning and some other adjustments e.g. in the RK4 integration to reduce register pressure, this could be made much faster, still.

Docker

docker run -it --gpus all --mount type=bind,source=$(pwd),target=/l2f-benchmark,readonly nvidia/cuda:12.6.2-devel-ubuntu24.04 bash
nvcc l2f-benchmark/benchmark.cu -I/l2f-benchmark/external/rl_tools/include -use_fast_math --optimize 3 -std=c++17 && ./a.out

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published