How to use anaconda tensorflow-gpu! Or how to use gpu about yolov5

[github-actions]

👋 Hello @rurusungoa, thank you for your interest in YOLOv5 🚀! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug it.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

Requirements

Python>=3.8.0 with all requirements.txt installed including PyTorch>=1.8. To get started:

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training, validation, inference, export and benchmarks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

Introducing YOLOv8 🚀

We're excited to announce the launch of our latest state-of-the-art (SOTA) object detection model for 2023 - YOLOv8 🚀!

Designed to be fast, accurate, and easy to use, YOLOv8 is an ideal choice for a wide range of object detection, image segmentation and image classification tasks. With YOLOv8, you'll be able to quickly and accurately detect objects in real-time, streamline your workflows, and achieve new levels of accuracy in your projects.

Check out our YOLOv8 Docs for details and get started with:

pip install ultralytics

[rurusungoa]

Could it be used in anaconda install?
I would like to inquire whether the GPU can be used with pytorch and cuda.

[glenn-jocher]

@rurusungoa hello! Thank you for reaching out with your question. 🌟

To use YOLOv5 with GPU acceleration, you don't need TensorFlow-GPU specifically, as YOLOv5 is built on PyTorch. To ensure GPU support, you should have a compatible version of PyTorch installed that works with CUDA on your system. This will allow YOLOv5 to leverage your GPU for training and inference tasks.

For Anaconda users, you can install PyTorch with CUDA support directly from the PyTorch official channel or through the conda-forge channel, ensuring that the CUDA version matches your system's CUDA installation. This setup is sufficient for running YOLOv5 on a GPU.

Regarding your setup with Red Hat OCP containers, as long as the container has access to a GPU and a compatible version of CUDA is installed, you should be able to use YOLOv5 with GPU acceleration without needing TensorFlow-GPU. Ensure that your container environment is properly configured to access the GPU, and you have installed the correct PyTorch and CUDA versions.

For more detailed instructions and requirements, please refer to our documentation at https://docs.ultralytics.com/yolov5/. This resource provides comprehensive guidance on setting up your environment for YOLOv5, including GPU support.

If you have any more questions or need further assistance, feel free to ask. Happy coding! 🚀

[rurusungoa]

This is the part where cuda was added as shown below. After installing the package, I check the graphics card and pytorch. If it is confirmed, how do I apply it to the py module contents and use it?

Package installation contents

(ultralytics) root@3ac25e0201ab:# conda list | grep cuda
cuda-version 11.8 h70ddcb2_2 conda-forge
cudatoolkit 11.8.0 h4ba93d1_13 conda-forge
(ultralytics) root@3ac25e0201ab:# conda list | grep cudnn
cudnn 8.8.0.121 hcdd5f01_4 conda-forge
(ultralytics) root@3ac25e0201ab:# conda list | grep pytorch
libopenvino-pytorch-frontend 2023.2.0 h59595ed_4 conda-forge
pytorch 2.1.0 cpu_mkl_py39h9c325db_103 conda-forge
(ultralytics) root@3ac25e0201ab:# conda list | grep opencv
libopencv 4.9.0 py39h0d0a14e_5 conda-forge
opencv 4.9.0 py39h49f0291_5 conda-forge
py-opencv 4.9.0 py39h0d77b2f_5 conda-forge
(ultralytics) root@3ac25e0201ab:~# conda list | grep ultraly

packages in environment at /opt/conda/envs/ultralytics:
ultralytics 8.1.15 pyh2965483_0 conda-forge

Check the drive after installing the package

import torch
torch.cuda.get_device_name(0) #gpu
torch.cuda.is_available() #cuda

Next, how do we apply it to the module and use it?

[glenn-jocher]

@rurusungoa hello again! It looks like you've successfully installed CUDA and PyTorch in your environment, which is great! To ensure that YOLOv5 utilizes your GPU, you generally don't need to make any manual changes to the code. YOLOv5 is designed to automatically detect and use available GPUs when running PyTorch with CUDA support.

From the package list you've provided, it seems you have installed pytorch 2.1.0 with CPU support (cpu_mkl). To leverage GPU acceleration, you'll need to install a version of PyTorch that supports CUDA. You can do this by installing the appropriate PyTorch version with CUDA support from the PyTorch channel in your conda environment. Make sure it matches the CUDA version you have installed (cuda-version 11.8 in your case).

Here's a general step you might follow to ensure your setup is correct for using a GPU with YOLOv5:

1. Install PyTorch with GPU Support: You need to replace the CPU version of PyTorch with a GPU-compatible version. You can find the correct command on the PyTorch website, selecting the configuration that matches your CUDA version. It typically looks something like this for CUDA 11.8 (make sure to adjust it according to the latest available versions and your specific setup):

   conda install pytorch torchvision torchaudio cudatoolkit=11.8 -c pytorch

2. Verify GPU Availability in PyTorch: After installing the correct version of PyTorch, you can verify that PyTorch recognizes your GPU using the commands you've mentioned:

   import torch
   print(torch.cuda.get_device_name(0))  # Should print the name of your GPU
   print(torch.cuda.is_available())  # Should return True if PyTorch can use the GPU

3. Running YOLOv5: When running YOLOv5 scripts such as training or detection, if PyTorch with CUDA support is correctly installed and configured, YOLOv5 will automatically use the GPU. There's no need for manual configuration in the module for GPU usage.

If you encounter any issues or have further questions, feel free to ask. Happy detecting! 🚀

[rurusungoa]

@glenn-jocher
YOLOv5 will automatically use the GPU. There's no need for manual configuration in the module for GPU usage.
=> Comment
What module does yolo version 5 use? When you say that it automatically uses the GPU, can you give me an example of using import ultralytics?

[glenn-jocher]

Hello @rurusungoa! I appreciate your follow-up question. To clarify, YOLOv5 is built on PyTorch, and it leverages PyTorch for all operations that can utilize the GPU. When I mention that YOLOv5 automatically uses the GPU, it means that the underlying PyTorch operations will default to using the GPU when it's available and properly configured with CUDA support.

Regarding your question about using import ultralytics, it seems there might be a slight misunderstanding. The typical way to use YOLOv5 in your code is not through an import ultralytics statement. Instead, you would interact with YOLOv5 through its scripts or by importing it as a module if you're integrating it into a larger Python project. The YOLOv5 repository itself is structured to be used directly or through cloning and running its scripts. For custom integration, you might directly use the models and utilities provided within the repository.

Here's a simple example of how you might load a YOLOv5 model and perform inference, assuming you have YOLOv5 cloned and set up in your environment:

import torch
from models.experimental import attempt_load  # This is how you might import the model loading function
from utils.general import non_max_suppression  # Example utility import

# Load the model
model = attempt_load('yolov5s.pt', map_location='cuda:0')  # Automatically uses GPU if available

# Perform inference
img = torch.zeros((1, 3, 640, 640)).cuda()  # Example input image tensor on GPU
pred = model(img)
pred = non_max_suppression(pred)  # Apply NMS

This example is quite simplified and assumes you're familiar with PyTorch and have the necessary YOLOv5 files. The key takeaway is that YOLOv5, through PyTorch, will automatically utilize the GPU if your environment is correctly set up with a CUDA-enabled version of PyTorch. There's no need for manual configuration specific to YOLOv5 to enable GPU usage.

For detailed examples and more comprehensive guidance on using YOLOv5, including loading models, performing inference, and training, please refer to the official documentation and examples provided in the YOLOv5 GitHub repository. If you have specific integration questions or need further assistance, feel free to ask. Happy coding! 🚀

[rurusungoa]

@glenn-jocher Thank you so much for glenn-jocher comment and interest.

When checking the GPU, I would like to inquire about the contents of import torch.

I installed the package with conda. Should I change “import torch” to “import pytorch” in the information you provided?

import torch
print(torch.cuda.get_device_name(0)) # Should print the name of your GPU
print(torch.cuda.is_available()) # Should return True if PyTorch can use the GPU

Need to search for changed content?

import pytorch
print(pytorch.cuda.get_device_name(0)) # Should print the name of your GPU
print(pytorch.cuda.is_available()) # Should return True if PyTorch can use the GPU

Please confirm.

[glenn-jocher]

Hello @rurusungoa! I'm here to clarify your question regarding the import statement for PyTorch.

You should use import torch and not import pytorch. The correct way to import PyTorch in your Python scripts remains import torch. The code snippets you've provided initially are correct for checking if your GPU is recognized by PyTorch and if PyTorch can utilize the GPU:

import torch
print(torch.cuda.get_device_name(0))  # Should print the name of your GPU
print(torch.cuda.is_available())      # Should return True if PyTorch can use the GPU

There's no need to change import torch to import pytorch. The package name for PyTorch in Python scripts is always torch, regardless of how it's installed (whether via pip, conda, or any other method).

So, to confirm, you should continue using import torch in your scripts when working with PyTorch. There's no "pytorch" module to import, and the information I provided earlier remains accurate.

If you have any more questions or need further assistance, feel free to ask. Keep up the great work! 🌟

[rurusungoa]

@glenn-jocher Thank you so much for glenn-jocher comment and interest.

thank you so much.
After applying it to jupyter-notebook and using it, once it is cleared, I will write again to express my gratitude.

[glenn-jocher]

You're very welcome, @rurusungoa! I'm glad I could assist you. Should you have any more questions or need further help as you work with YOLOv5 in your Jupyter Notebook, please don't hesitate to reach out. Wishing you the best of luck with your project, and I look forward to hearing about your progress. Happy coding! 🚀

[rurusungoa]

@glenn-jocher I'm sorry for keeping asking questions.

I checked the pytorch version by installing pytorch, cudnn, and cudatoolkit separately as shown below.
"AssertionError: Torch not comiled with CUDA enabled" It doesn't seem to be available.

(ultralytics) root@9c6698432b05:~# conda list | grep ult
libclang 15.0.7 default_hb11cfb5_4 conda-forge
libclang13 15.0.7 default_ha2b6cf4_4 conda-forge
ultralytics 8.1.16 pyh2965483_0 conda-forge

(ultralytics) root@9c6698432b05:~# conda list | grep opencv
libopencv 4.7.0 py38hb32f1de_1 conda-forge
opencv 4.7.0 py38h578d9bd_1 conda-forge
opencv-python-headless 4.9.0.80 py38_0 fastai
py-opencv 4.7.0 py38h6f1a3b6_1 conda-forge

(ultralytics) root@9c6698432b05:~# conda list | grep torc
libopenvino-pytorch-frontend 2023.2.0 h59595ed_4 conda-forge
libtorch 2.1.0 cpu_mkl_hadc400e_103 conda-forge
pytorch 2.1.0 cpu_mkl_py39h9c325db_103 conda-forge
torchdata 0.7.1 py39h881704f_1 conda-forge
torchtext 0.15.2 py39hd75fb19_4 conda-forge
torchvision 0.16.1 cpu_py39h7f5b5c8_2 conda-forge

(ultralytics) root@9c6698432b05:~# conda list | grep cuda
cuda-version 11.8 h70ddcb2_2 conda-forge
cudatoolkit 11.8.0 h4ba93d1_13 conda-forge

(ultralytics) root@9c6698432b05:~# conda list | grep cudnn
cudnn 8.8.0.121 hcdd5f01_4 conda-forge

So, I'm going to use the information you gave me last time to create a pytorch channel and install cudnn.

conda install pytorch torchvision torchaudio cudatoolkit=11.8 -c pytorch -y

(python38) root@9c6698432b05:~# conda list | grep pytorch
libopenvino-pytorch-frontend 2023.3.0 h59595ed_3 conda-forge
pytorch 2.1.2 cpu_mkl_py38h8efc972_100 conda-forge
pytorch-mutex 1.0 cpu pytorch
torchaudio 2.1.2 py38_cpu pytorch
torchvision 0.16.2 py38_cpu pytorch

(python38) root@9c6698432b05:~# conda list | grep cuda
cudatoolkit 11.8.0 h4ba93d1_13 conda-forge

Is it okay that pytorch-cuda is not installed here? Can I use it like this?

[rurusungoa]

Hello @glenn-jocher I'm sorry for keeping asking questions.
I sincerely ask for your answer.

• gpu test in jupyterlab

1. device check

import torch
print(torch.cuda.get_device_name(0))

--------------------------------------------------------------------------- AssertionError Traceback (most recent call last)
Cell In[5], line 1 ----> 1 torch.cuda.get_device_name(0)

File /opt/conda/envs/python310/lib/python3.10/site-packages/torch/cuda/init.py:419, in get_device_name(device) 407 def get_device_name(device: Optional[_device_t] = None) -> str:
408 r"""Gets the name of a device. 409 410 Args: (...) 417 str: the name of the device 418 """ --> 419 return get_device_properties(device).name

File /opt/conda/envs/python310/lib/python3.10/site-packages/torch/cuda/init.py:449, in get_device_properties(device) 439 def get_device_properties(device: _device_t) -> _CudaDeviceProperties:
440 r"""Gets the properties of a device. 441 442 Args: (...) 447 _CudaDeviceProperties: the properties of the device 448 """ --> 449 _lazy_init() # will define _get_device_properties 450 device = _get_device_index(device, optional=True)
451 if device < 0 or device >= device_count():

File /opt/conda/envs/python310/lib/python3.10/site-packages/torch/cuda/init.py:289, in _lazy_init() 284 raise RuntimeError(
285 "Cannot re-initialize CUDA in forked subprocess. To use CUDA with " 286 "multiprocessing, you must use the 'spawn' start method" 287 )
288 if not hasattr(torch._C, "_cuda_getDeviceCount"):
--> 289 raise AssertionError("Torch not compiled with CUDA enabled")
290 if _cudart is None:
291 raise AssertionError(
292 "libcudart functions unavailable. It looks like you have a broken build?" 293 )

AssertionError: Torch not compiled with CUDA enabled

2. Check whether it can be used with cuda

print(torch.cuda.is_available())
False

3. List of installed packages

The python version is 3.10.14. Since I couldn't install it with cudatoolkit, I tried using pytorch-cuda as shown below.
But the same error as cudatoolkit occurs.

conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=11.8 -c pytorch -c nvidia

conda list | grep python

python 3.10.14 hd12c33a_0_cpython conda-forge

conda list | grep pytorch

libopenvino-pytorch-frontend 2023.3.0 h59595ed_3 conda-forge
pytorch 2.1.2 cpu_mkl_py310h3ea73d3_100 conda-forge
pytorch-cuda 12.1 ha16c6d3_5 pytorch
pytorch-mutex 1.0 cuda pytorch
torchaudio 2.1.2 py310_cu121 pytorch

conda list | grep torch

libopenvino-pytorch-frontend 2023.3.0 h59595ed_3 conda-forge
libtorch 2.1.2 cpu_mkl_hadc400e_100 conda-forge
pytorch 2.1.2 cpu_mkl_py310h3ea73d3_100 conda-forge
pytorch-cuda 12.1 ha16c6d3_5 pytorch
pytorch-mutex 1.0 cuda pytorch
torchaudio 2.1.2 py310_cu121 pytorch
torchdata 0.7.1 py310h0bd2ee8_3 conda-forge
torchtext 0.15.2 py310h4e894d6_4 conda-forge
torchvision 0.16.1 cpu_py310h684a773_3 conda-forge

conda list | grep cuda

cuda-cudart 12.1.105 0 nvidia
cuda-cupti 12.1.105 0 nvidia
cuda-libraries 12.1.0 0 nvidia
cuda-nvrtc 12.1.105 0 nvidia
cuda-nvtx 12.1.105 0 nvidia
cuda-opencl 12.4.99 h59595ed_0 conda-forge
cuda-runtime 12.1.0 0 nvidia
cuda-version 12.4 h3060b56_3 conda-forge
pytorch-cuda 12.1 ha16c6d3_5 pytorch
pytorch-mutex 1.0 cuda pytorch

• nvidia-smi driver version NVIDIA-SMI 470.82.01 is installed.

I installed the package as above, but can you check the error message? Please. I don't know why it doesn't work. help plz..(T.T) It's so sad

[glenn-jocher]

Hello @rurusungoa, no need to apologize for asking questions; we're here to help!

From the details you've provided, it appears that PyTorch has been installed with CPU support only, as indicated by cpu_mkl in the version of PyTorch listed (pytorch 2.1.2 cpu_mkl_py310h3ea73d3_100). This is why you're encountering the "Torch not compiled with CUDA enabled" error and why torch.cuda.is_available() returns False.

To resolve this, you'll need to ensure that you install the GPU version of PyTorch. The Conda command you used seems correct, but it looks like it might have defaulted to installing the CPU version, possibly due to package constraints or compatibility issues with the specified packages.

Try the following steps to install the appropriate versions that support CUDA:

1. Make sure your Conda environment is activated.
2. Use the command you provided, which is generally correct, but double-check the PyTorch, CUDA, and NVIDIA channels for the latest versions and compatibility:

conda install pytorch torchvision torchaudio cudatoolkit=11.8 -c pytorch -c nvidia

3. After installation, you can verify that a GPU-supporting version of PyTorch is installed by checking the packages list again or by running torch.cuda.is_available() in a Python shell or Jupyter notebook, which should return True if everything is set up correctly.

Regarding the pytorch-cuda package you mentioned, that's not a typical package name for installing PyTorch with GPU support. The correct package for enabling GPU support in PyTorch when using Conda is cudatoolkit, which you've included in your command.

If these steps do not resolve the issue, consider creating a fresh Conda environment before reinstalling the packages. Sometimes, package constraints from existing installations can prevent the correct versions from being installed.

Keep up the perseverance, and again, feel free to ask more questions if you need further assistance. 🚀

[rurusungoa]

Hello @glenn-jocher

Make sure your Conda environment is activated.
Use the command you provided, which is generally correct, but double-check the PyTorch, CUDA, and NVIDIA channels for the latest versions and compatibility:
=> answer

When I checked the pytorch conda homepage, the following information appeared, so I proceeded with the installation.

When I checked on the pytorch homepage

https://pytorch.org/get-started/previous-versions/
v2.1.2
conda
conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=11.8 -c pytorch -c nvidia

The pytorch driver is installed as shown below.

pytorch 2.1.2 cpu_mkl_py310h3ea73d3_100 conda-forge

Install the image package on a regular PC
Next, I checked this by accessing a server with a GPU.
Do I need to perform CONDA INSTALL on a server with a GPU from the beginning?

[glenn-jocher]

Hello @rurusungoa,

Yes, for PyTorch to properly utilize GPU capabilities, it's essential that the GPU-supported version of PyTorch is installed on the machine that has the GPU. Since the version installed (pytorch 2.1.2 cpu_mkl_py310h3ea73d3_100 conda-forge) is CPU-only, you'll need to reinstall PyTorch with GPU support on your server.

The command you found on the PyTorch website is generally correct, but the pytorch-cuda=11.8 part seems to be a misunderstanding. To install the GPU version, you only need to specify the cudatoolkit version compatible with your GPU drivers. Here's a simplified command:

conda install pytorch torchvision torchaudio cudatoolkit=11.8 -c pytorch

Ensure this installation is performed on the server where the GPU is present, and after installation, running torch.cuda.is_available() in Python should return True, indicating that PyTorch can now utilize the GPU.

Happy coding! 🚀

[rurusungoa]

Hello @glenn-jocher

As I mentioned, I installed pytorch in the GPU environment. The installation continues with CPU allocation.

pytorch 2.1.2 cpu_mkl_py310h3ea73d3_100 conda forge

conda install pytorch torchvision torchaudio cudatoolkit=11.8 -c pytorch

What is the cause? Is there a reason why it can't be installed?

Run watch on the system

watch -n 1 -d nvidia-smi

Could this be done by going to conda virtual in podman images and installing it?

Should I install it with pip file and then install it with conda?

Is there a Docker image file used by jupyter-lab?

[glenn-jocher]

@rurusungoa hello! It seems like you're encountering persistent issues with installing the GPU version of PyTorch. The installation command you're using looks correct for installing the GPU-supported PyTorch, so here are a few quick suggestions:

1. Ensure you're executing the installation in the right environment where the GPU is accessible. Using a system monitor (nvidia-smi) to check GPU usage won't directly impact the PyTorch installation but is good for verifying that your GPU is recognized by the system.

2. Installing PyTorch within a container environment (like Docker or Podman) requires the base image to have CUDA support. For JupyterLab, you might want to start with a CUDA-enabled base image from NVIDIA and install your packages on top of that.

3. There's no need to install with a pip file first. The Conda installation should suffice. However, make sure that your environment doesn't have any conflicting packages that might force PyTorch to fall back to a CPU version.

4. If you're using Docker, NVIDIA provides PyTorch images pre-configured with CUDA, cuDNN, and other dependencies on NVIDIA NGC. These can save you some setup time.

Double-check the CUDA compatibility with your GPU and the PyTorch version you're trying to install. Sometimes picking a specific cudatoolkit version that matches your system's CUDA version can help.

If issues persist, creating a fresh environment and reinstalling might help clear any unseen conflicts. Keep at it, and let us know if there's more we can do to assist! 🚀

[rurusungoa]

Hello @glenn-jocher

Do you have any docker images with cudatoolkit=11.8?
Are there any images I can "docker pull" and use?
If you provide it, I would like to test it.

[glenn-jocher]

@rurusungoa hello! 😊

We don't offer official Ultralytics Docker images with specific cudatoolkit=11.8 versions. However, you can easily create your own Docker image based on NVIDIA's PyTorch images, which come with CUDA support. Here's a simple Dockerfile example to get you started:

FROM pytorch/pytorch:1.8.1-cuda11.1-cudnn8-runtime

RUN pip install ultralytics

You can adjust the base image to match the cudatoolkit=11.8 version you're interested in. After setting up your Dockerfile, build and run your Docker image to get started with your tests.

Happy testing! Let us know if you have more questions.

[rurusungoa]

Hello @glenn-jocher

After installing and connecting the NVIDIA-GTX-1060 graphics card, I ran docker run --runtime to check the GPU.

After that, I installed pytorch and checked the device, but the result was "false".
I looked at the CPU device.

So, I checked the GPU device with the image through the link below and it was successful and true.
docker pull pytorch/pytorch:2.2.2-cuda11.8-cudnn8-runtime

This time I read the GPU card fine. It was a success.

But I cannot access jupyter-lab. What should I do?
pip uninstall jupyter-related packages and
Even if you install conda install jupyter related packages, access is not possible.
Even with CMD jupyter lab --allow-root, it seems like I can't access the web.
Do you know how to initialize jupyter-notebook?

Can you tell me how to approach this?

[glenn-jocher]

Hello @rurusungoa,

Great to hear you've successfully accessed your GPU within the Docker container! 😊 For JupyterLab access issues, it often comes down to port forwarding. When you run a JupyterLab server inside a Docker container, you need to map the container's port to a port on your host machine.

With Docker, you can add the -p option to your docker run command to specify port mappings. If your JupyterLab server inside the container runs on the default port 8888, you can map this to the same port or another available port on your host machine. Here’s how you could do it:

docker run -p 8888:8888 --runtime=nvidia YOUR_IMAGE_NAME jupyter lab --allow-root --ip=0.0.0.0 --NoBrowser

This command maps port 8888 inside the container to port 8888 on your host, allowing you to access JupyterLab by visiting http://localhost:8888 in your web browser. Also, make sure JupyterLab is installed inside your Docker image. If not, include it in your Dockerfile:

RUN pip install jupyterlab

Build your Docker image with this requirement, and try running it again with the port forwarding. Hope this helps you get JupyterLab up and running!

[rurusungoa]

Hello @glenn-jocher

glenn-jocher Thank you. Thank you very much.
Well resolved.
We are using redhat software "openshift" environment.
As you mentioned, the GPU device was loaded successfully using the pytorch/pytorch:1.8.1-cuda11.1-cudnn8-runtime image.
thank you

Even though I did not add the nvidia gpu(nvidia.com/gpu:1) device to the "deployment" in openshfit,
The nvidia device was loaded.

I don't understand that part, but it seems like there is some kind of environment setting, and I don't understand it, but it's going well.

[glenn-jocher]

Hello @rurusungoa,

I'm thrilled to hear that your issue has been resolved and that the GPU device is successfully loaded in your OpenShift environment using the specified PyTorch image! 🎉 Sometimes, configurations within container orchestration environments like OpenShift may automatically detect and utilize available hardware resources such as GPUs, depending on the underlying setup and node configurations. It's possible that your OpenShift cluster is configured to make GPUs available to pods without explicitly specifying them in the deployment configuration.

This "auto-magic" behavior can be convenient, but for best practices and clarity, I'd recommend reviewing your OpenShift cluster's documentation or configurations related to GPU resources. This will give you a clearer understanding and more predictable control over how resources are allocated to your pods.

Keep up the great work, and don't hesitate to reach out if you have more questions or run into new challenges. Happy coding! 🚀