Skip to content

segmentation fault——CUDA 12.8 for RTX 50 series support #14

@1nibeloflight2

Description

@1nibeloflight2

Hello, I’ve been trying to run the OneMap evaluation with an RTX 5090 GPU, but I keep encountering intermittent segmentation faults during execution. The error occurs randomly in the middle of evaluation, preventing me from completing a full run.
I’m wondering if this instability could be related to the PyTorch version I’m using: torch 2.11.0+cu128, which is required to support the RTX 5090.
Error logs are as follows:

(onemap) ➜  OneMap git:(main) ✗ python eval_habitat.py --config config/mon/eval_conf.yaml
anaconda3/envs/onemap/lib/python3.10/site-packages/timm/models/layers/__init__.py:49: FutureWarning: Importing from timm.models.layers is deprecated, please import via timm.layers
  warnings.warn(f"Importing from {__name__} is deprecated, please import via timm.layers", FutureWarning)
TensorRT not available, cannot use Jetson
[1]    90148 segmentation fault (core dumped)  python eval_habitat.py --config config/mon/eval_conf.yaml

Or sometimes it crashes later during inference:


YOLO forward: 0.012259244918823242
YOLO forward: 0.012404680252075195
YOLO forward: 0.01256561279296875
YOLO forward: 0.012644529342651367
tensor(0.8005, device='cuda:0')
YOLO forward: 0.01303863525390625
[1]    100698 segmentation fault (core dumped)  python3 eval_habitat.py --config config/mon/eval_conf.yaml

Any insights or suggestions would be greatly appreciated. Thank you!

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