Skip to content

IsaacGym crashes with 2048 parallel environments on H800 when training h1_2 mimic task #222

Description

@zhang705

When running the mimic training task for the h1_2 robot on NVIDIA H800 with --num-envs 2048, IsaacGym crashes immediately with GPU PhysX internal errors and segmentation fault.

The code works with smaller numbers of environments (1024), but crashes with just num-envs=2048

Command:
CUDA_VISIBLE_DEVICES=7 python protomotions/train_agent.py
--robot-name h1_2
--simulator isaacgym
--experiment-path examples/experiments/mimic/mlp.py
--experiment-name mimic
--motion-file data/motion_for_trackers/h1_2_random_subset_tiny.pt
--num-envs 2048
--batch-size 2048
--ngpu 1

Error Log:

/buildAgent/work/99bede84aa0a52c2/source/gpucommon/include/PxgCudaUtils.h (54) : internal error : SynchronizeStreams cuEventRecord failed

/buildAgent/work/99bede84aa0a52c2/source/gpucommon/include/PxgCudaUtils.h (60) : internal error : SynchronizeStreams cuStreamWaitEvent failed

/buildAgent/work/99bede84aa0a52c2/source/gpunarrowphase/src/PxgNarrowphaseCore.cpp (2408) : internal error : memcpy failed fail!
700

/buildAgent/work/99bede84aa0a52c2/source/gpucommon/src/PxgCudaMemoryAllocator.cpp (59) : warning : Failed to allocate pinned memory.

Segmentation fault (core dumped)

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions