I want to use Docker to run an image to utilize the GPU on Jetson, but I found that the GPU is not being used

Robotics & Edge Computing Jetson & Embedded Systems Jetson Orin Nano
1

my jetson orin nano jetpack 6.2.1

my docker version
root@nvidia-desktop:/home/nvidia# docker version
Client: Docker Engine - Community
Version: 28.3.2
API version: 1.51
Go version: go1.24.5
Git commit: 578ccf6
Built: Wed Jul 9 16:13:42 2025
OS/Arch: linux/arm64
Context: default

Server: Docker Engine - Community
Engine:
Version: 28.3.2
API version: 1.51 (minimum version 1.24)
Go version: go1.24.5
Git commit: e77ff99
Built: Wed Jul 9 16:13:42 2025
OS/Arch: linux/arm64
Experimental: true
containerd:
Version: 1.7.27
GitCommit: 05044ec0a9a75232cad458027ca83437aae3f4da
nvidia:
Version: 1.2.5
GitCommit: v1.2.5-0-g59923ef
docker-init:
Version: 0.19.0
GitCommit: de40ad0

NVIDIA Container Toolkit

root@nvidia-desktop:/home/nvidia# nvidia-container-runtime --version
NVIDIA Container Runtime version 1.17.8
commit: f202b80a9b9d0db00d9b1d73c0128c8962c55f4d
spec: 1.2.1

runc version 1.2.5
commit: v1.2.5-0-g59923ef
spec: 1.2.0
go: go1.23.7
libseccomp: 2.5.3
root@nvidia-desktop:/home/nvidia# docker info | grep Runtimes
Runtimes: io.containerd.runc.v2 nvidia runc
root@nvidia-desktop:/home/nvidia#

docker daemon.json
root@nvidia-desktop:/home/nvidia# cat /etc/docker/daemon.json
{
“builder”: {
“gc”: {
“defaultKeepStorage”: “20GB”,
“enabled”: true
}
},
“data-root”: “/var/lib/docker”,
“exec-opts”: [
“native.cgroupdriver=systemd”
],
“experimental”: true,
“features”: {
“buildkit”: true
},
“log-driver”: “json-file”,
“log-opts”: {
“max-file”: “60”,
“max-size”: “500m”
},
“registry-mirrors”: [
“https://ghcr.ikubernetes.xyz”,
“https://cloudsmith.ikubernetes.xyz”,
“https://docker.ikubernetes.xyz”,
“https://quey.ikubernetes.xyz”,
“https://gcr.ikubernetes.xyz”,
“https://k8s-gcr.ikubernetes.xyz”,
“https://docker.registry.cyou”,
“https://docker-cf.registry.cyou”,
“https://dockercf.jsdelivr.fyi”,
“https://docker.jsdelivr.fyi”,
“https://dockertest.jsdelivr.fyi”,
“https://mirror.aliyuncs.com”,
“https://dockerproxy.com”,
“https://mirror.baidubce.com”,
“https://docker.m.daocloud.io”,
“https://docker.nju.edu.cn”,
“https://docker.mirrors.sjtug.sjtu.edu.cn”,
“https://docker.mirrors.ustc.edu.cn”,
“https://mirror.iscas.ac.cn”,
“https://docker.rainbond.cc”,
“https://k8s..ikubernetes.xyz”
],
“runtimes”: {
“nvidia”: {
“args”: ,
“path”: “nvidia-container-runtime”,
“runtimesArgs” : 

    }
},
"default-runtime":"nvidia"

}
root@nvidia-desktop:/home/nvidia#

Docker Test the jetsonGPU
PyTorch training code mnist.py

from future import print_function
import argparse
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
from torch.optim.lr_scheduler import StepLR

class Net(nn.Module):
def init(self):
super(Net, self).init()
self.conv1 = nn.Conv2d(1, 32, 3, 1)
self.conv2 = nn.Conv2d(32, 64, 3, 1)
self.dropout1 = nn.Dropout(0.25)
self.dropout2 = nn.Dropout(0.5)
self.fc1 = nn.Linear(9216, 128)
self.fc2 = nn.Linear(128, 10)

def forward(self, x):
    x = self.conv1(x)
    x = F.relu(x)
    x = self.conv2(x)
    x = F.relu(x)
    x = F.max_pool2d(x, 2)
    x = self.dropout1(x)
    x = torch.flatten(x, 1)
    x = self.fc1(x)
    x = F.relu(x)
    x = self.dropout2(x)
    x = self.fc2(x)
    output = F.log_softmax(x, dim=1)
    return output

def train(args, model, device, train_loader, optimizer, epoch):
model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target.to(device)
optimizer.zero_grad()
output = model(data)
loss = F.nll_loss(output, target)
loss.backward()
optimizer.step()
if batch_idx % args.log_interval == 0:
print(‘Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}’.format(
epoch, batch_idx * len(data), len(train_loader.dataset),
100. * batch_idx / len(train_loader), loss.item()))
if args.dry_run:
break

def test(model, device, test_loader):
model.eval()
test_loss = 0
correct = 0
with torch.no_grad():
for data, target in test_loader:
data, target = data.to(device), target.to(device)
output = model(data)
test_loss += F.nll_loss(output, target, reduction=‘sum’).item() # sum up batch loss
pred = output.argmax(dim=1, keepdim=True) # get the index of the max log-probability
correct += pred.eq(target.view_as(pred)).sum().item()

test_loss /= len(test_loader.dataset)

print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
    test_loss, correct, len(test_loader.dataset),
    100. * correct / len(test_loader.dataset)))

def main():
# Training settings
parser = argparse.ArgumentParser(description=‘PyTorch MNIST Example’)
parser.add_argument(‘–batch-size’, type=int, default=64, metavar=‘N’,
help=‘input batch size for training (default: 64)’)
parser.add_argument(‘–test-batch-size’, type=int, default=1000, metavar=‘N’,
help=‘input batch size for testing (default: 1000)’)
parser.add_argument(‘–epochs’, type=int, default=14, metavar=‘N’,
help=‘number of epochs to train (default: 14)’)
parser.add_argument(‘–lr’, type=float, default=1.0, metavar=‘LR’,
help=‘learning rate (default: 1.0)’)
parser.add_argument(‘–gamma’, type=float, default=0.7, metavar=‘M’,
help=‘Learning rate step gamma (default: 0.7)’)
parser.add_argument(‘–no-cuda’, action=‘store_true’, default=False,
help=‘disables CUDA training’)
parser.add_argument(‘–no-mps’, action=‘store_true’, default=False,
help=‘disables macOS GPU training’)
parser.add_argument(‘–dry-run’, action=‘store_true’, default=False,
help=‘quickly check a single pass’)
parser.add_argument(‘–seed’, type=int, default=1, metavar=‘S’,
help=‘random seed (default: 1)’)
parser.add_argument(‘–log-interval’, type=int, default=10, metavar=‘N’,
help=‘how many batches to wait before logging training status’)
parser.add_argument(‘–save-model’, action=‘store_true’, default=False,
help=‘For Saving the current Model’)
args = parser.parse_args()
use_cuda = not args.no_cuda and torch.cuda.is_available()
use_mps = not args.no_mps and torch.backends.mps.is_available()

torch.manual_seed(args.seed)

if use_cuda:
    device = torch.device("cuda")
elif use_mps:
    device = torch.device("mps")
else:
    device = torch.device("cpu")

train_kwargs = {'batch_size': args.batch_size}
test_kwargs = {'batch_size': args.test_batch_size}
if use_cuda:
    cuda_kwargs = {'num_workers': 1,
                   'pin_memory': True,
                   'shuffle': True}
    train_kwargs.update(cuda_kwargs)
    test_kwargs.update(cuda_kwargs)

transform=transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize((0.1307,), (0.3081,))
    ])
dataset1 = datasets.MNIST('../data', train=True, download=True,
                   transform=transform)
dataset2 = datasets.MNIST('../data', train=False,
                   transform=transform)
train_loader = torch.utils.data.DataLoader(dataset1,**train_kwargs)
test_loader = torch.utils.data.DataLoader(dataset2, **test_kwargs)

model = Net().to(device)
optimizer = optim.Adadelta(model.parameters(), lr=args.lr)

scheduler = StepLR(optimizer, step_size=1, gamma=args.gamma)
for epoch in range(1, args.epochs + 1):
    train(args, model, device, train_loader, optimizer, epoch)
    test(model, device, test_loader)
    scheduler.step()

if args.save_model:
    torch.save(model.state_dict(), "mnist_cnn.pt")

if name == ‘main’:
main()

Create a Docker image using a Dockerfile, based on the NVIDIA PyTorch base image.
FROM nvcr.io/nvidia/l4t-pytorch:r35.2.1-pth2.0-py3

COPY pytorch-mnist.py /home/

docker build -t mnist:1.0 .

Create a Docker image using a Dockerfile, based on the NVIDIA PyTorch base image.

docker run -it --runtime nvidia mnist:1.0 /bin/bash

python3 /home/pytorch-minst.py

Find GPU not be used，CPU be used.

When I run mnist.py directly on Jetson, I find that it can use the GPU.

I don’t know where I made the mistake. Can you help me?

3

Hi,

It looks like you are using the r35.2.1 image (from JetPack 5).
Which JetPack version do you use for setting up the device?

Thanks.