Skip to content

environment (important)

system

Instances are run as Docker containers, and the system is unified as Ubuntu 18.04. There are the following restrictions on container instances:

  • The restart command cannot be used, please restart the instance through the console
  • The kernel version and graphics card driver cannot be upgraded or upgraded
  • Cannot install and use Docker containers
  • After the instance is created, the / space capacity is 20G, please store the data in the /hy-tmp/ directory

To shutdown, you can use the shutdown command, a prompt will appear after execution, wait for a few seconds to complete the shutdown.

resource

The number of processor cores and memory of an instance is calculated based on the ratio of rented graphics cards to all graphics cards in the machine. The cat /proc/cpuinfo and free commands get all the resources on the machine, not the resources available to the instance. Instance resources can be viewed using the following commands.

# Check the number of processor cores
echo $(($(cat /sys/fs/cgroup/cpu/cpu.cfs_quota_us) / $(cat /sys/fs/cgroup/cpu/cpu.cfs_period_us)))

# View instance memory
echo "$(($(cat /sys/fs/cgroup/memory/memory.limit_in_bytes) / 1024 / 1024 / 1024)) GB"

# View the instance root partition disk
df -h | grep "/$" | awk '{print $5" "$3"/"$2}'

The internet

The instance does not have a public IP, and the network service with monitoring enabled in the instance cannot be accessed through the access point (i-N.gpushare.com). SSH, JupyterLab and TensorBoard are accessed through the access point through port mapping.

graphics card

Due to container limitations, using nvidia-smi cannot see which process is using the card. You can use the py3smi command to view (Owner, PID and other information will be inaccurate).

CUDA Version in nvidia-smi indicates the highest CUDA version supported by the driver, not the CUDA version in the instance. You can use the following commands to view CUDA related information.

# View CUDA version
nvcc -V

# View CUDNN version
dpkg -l | grep -E "\slibcudnn[0-9]\s" | awk '{print $2": "$3}'

# View CUDNN location
dpkg -L $(dpkg -l | grep -E "\slibcudnn[0-9]\s" | awk '{print $2}')

# View NCCL locations
dpkg -L libnccl2

Python

warning

It is not supported to delete or uninstall the built-in Python of the system in any way, otherwise the instance will no longer be able to start normally.

Python is pre-installed in the instance, and the Python version is selected when creating the official image of the instance. Libraries and frameworks are installed with this Python version.

Both python and python3 are soft links, linking to the built-in Python version of the mirror.

# View Python version
python -V

# Soft links for Python commands
ll /usr/local/bin/python*