Skip to main content
Docker is not supported within the container instance.

​
About System Disk

The instance’s system disk and data disk are generally local SSDs, with a small number being cloud disks. Local data disks offer good performance but lack redundant copies. Due to the inherent lifespan limitations of local disks, there is a probability of failure. Important data should be regularly backed up to file storage or locally. Besides, GPUhub does not provide reliability guarantees or services for such storage products.

​
Instance GPU/CPU/Memory Allocation

In GPUhub the method for renting GPUs involves selecting a physical host and then creating an instance with N GPU cards on that host. The allocation logic for memory and CPU configurations is based on a multiple of the number of GPUs. For example, if the CPU and memory allocation rules on the host are 32GB/GPU and 8 cores/GPU, then:
  • Renting 1 GPU would result in an instance configuration of 1 GPU, 8 CPU cores, and 32GB of memory.
  • Renting 2 GPUs would result in an instance configuration of 2 GPUs, 16 CPU cores, and 64GB of memory,
  • and so on.
Therefore, when you rent an instance, you should choose an appropriate physical host based on your needs and then select the number of GPUs required.

​
Directories within the Instance

To check the disk usage within the instance, please execute the command source /root/.bashrc in the terminal.

​
Further Explanation of the System Disk

In the Linux operating system, almost everything exists in the form of folders or files, and these folders/files share a common root directory /. If we install the Linux operating system on a disk A (with no other partitions), the root directory will contain the following default system directories:
bash
By default, writing files to any directory ultimately writes to disk A. In an GPUhub instance, the system disk refers to this disk A where the system is installed. Writing files to any location under the root directory / means writing to the system disk. However, there are exceptions.
  • For example, if there is another new disk B and we want to use it in the Linux system, we can mount this disk device to a specific directory (assuming the mount directory is /root/gpuhub-tmp).
In this case, writing “folders/files” to the /root/gpuhub-tmp directory means writing to disk B, not disk A. Here, you should understand the boundary between the system disk and the data disk in GPUhub.
  • For example, system software installed in GPUhub instances and Python dependencies installed using the built-in Miniconda are all installed on the system disk.

​
JupyterLab Working Directory

The working path of JupyterLab is /root. The data disk, network disk, and public data directory are all located under the /root directory. Therefore, when you enter JupyterLab, you can directly see these directories created by the platform by default. If you need to view the root directory /, you can enter the terminal and use command-line operations.