High CPU usage during evaluation

[opendrivelab]

High CPU usage during evaluation

[sqb2145]

Hi,

Can you please help me with running the run_evaluation.sh file. Like how you edited the file and where you placed the ckpt file.

Thank You!

[penghao-wu]

Hi, your 2.3G GPU memory usage is expected. In my case, the python script has a ~180% CPU usage and the CarlaUE4 has a ~200% CPU usage.

[HanaRo]

Hi, your 2.3G GPU memory usage is expected. In my case, the python script has a ~180% CPU usage and the CarlaUE4 has a ~200% CPU usage.

@WPH-commit
Thanks for the information!
But still I'm confused about why it utilizes so much CPU rather than GPU in my case. Since my evaluation in running on a shared server, it's kind of unaffordable for a single process using up the CPU.
Please let me know if I need to provide more information. Thx!

[HanaRo]

Hi,

Can you please help me with running the run_evaluation.sh file. Like how you edited the file and where you placed the ckpt file.

Thank You!

@sqb2145
In my case, I edited the 'CARLA_ROOT' and 'TEAM_CONFIG' in the script.
For the ckpt file, I put it at '${TCP_ROOT}/xxx/xx.ckpt', so the 'TEAM_CONFIG' is edited to 'xxx/xx.ckpt'
Hope this will help you :)

[penghao-wu]

But still I'm confused about why it utilizes so much CPU rather than GPU in my case. Since my evaluation in running on a shared server, it's kind of unaffordable for a single process using up the CPU.

Could you check the GPU usage for the Carla server? It should be around 1G.

[HanaRo]

But still I'm confused about why it utilizes so much CPU rather than GPU in my case. Since my evaluation in running on a shared server, it's kind of unaffordable for a single process using up the CPU.

Could you check the GPU usage for the Carla server? It should be around 1G.

@WPH-commit
Yes CARLA server basically run as you said, with about 1G memory, 200% CPU and about 30% GPU.

[sqb2145]

Hi,
Can you please help me with running the run_evaluation.sh file. Like how you edited the file and where you placed the ckpt file.
Thank You!

@sqb2145 In my case, I edited the 'CARLA_ROOT' and 'TEAM_CONFIG' in the script. For the ckpt file, I put it at '${TCP_ROOT}/xxx/xx.ckpt', so the 'TEAM_CONFIG' is edited to 'xxx/xx.ckpt' Hope this will help you :)

Hi,
Thank you for your reply. I think the issue I'm having isn't because of placing the ckpt file in the wrong directory. I'll try to figure it out. Thanks a lot! :)

[Naive-Bayes]

Hi, your 2.3G GPU memory usage is expected. In my case, the python script has a ~180% CPU usage and the CarlaUE4 has a ~200% CPU usage.

@WPH-commit Thanks for the information! But still I'm confused about why it utilizes so much CPU rather than GPU in my case. Since my evaluation in running on a shared server, it's kind of unaffordable for a single process using up the CPU. Please let me know if I need to provide more information. Thx!

I think this problem is not because the model, but the Carla Simulator. When do the open-loop training, it really faster than eval. But when we do close-loop evaluation, it is so slow. Actually, I also do not have the solution to accelerate the simulation process，sad.