PPO experiment not using GPU

RLlib
1

How severe does this issue affect your experience of using Ray?

  • Medium: It contributes to significant difficulty to complete my task, but I can work around it.

OS: Ubuntu 18.04.6 LTS
GPU: Nvidia GeForce RTX3060
ray version 2.6.3
ray-cpp version 2.1.0
python 3.8.15

I have two experiments one using SAC and another using PPO, the file that starts the experiment is identical apart form the algorithm itself and goes something as follows

ray.init( num_gpus=1,include_dashboard=True,_temp_dir="/bla/bla/ray_logs")
analysis = tune.run(CustomPPOTrainer,
                 name=args.name,
                 local_dir=args.directory,
                 # stop={"perf/ram_util_percent": 85.0},
                 checkpoint_freq=500,
                 # checkpoint_at_end=True,
                 restore=get_checkpoint(args.name, args.directory, args.restore, args.overwrite),
                 config=args.config,
                 # queue_trials=True,
                 resume=False,
                 reuse_actors=True,
                 )

Furthermore, in each of their config they both have

num_gpus: 1

The issue is that when I run the SAC experiment, it seems to be using the GPU displaying this message

Trial status: 1 PENDING
Current time: 2023-09-19 19:48:08. Total running time: 0s
Logical resource usage: 15.0/20 CPUs, 1.0/1 GPUs (0.0/1.0 accelerator_type:G)

While when running the PPO experiment, the following is displayed

Trial status: 1 PENDING
Current time: 2023-09-19 19:50:56. Total running time: 6s
Logical resource usage: 19.0/20 CPUs, 0/1 GPUs (0.0/1.0 accelerator_type:G)

What could be the cause that the PPO experiment doesn’t make use of the GPU while the SAC does.

Thank you in advance for your help.