Skip Ribbon Commands
Skip to main content

MLDA@EEE GPU Resource Management and Policy

  1. Resources

    GPU Server with 4 x NVIDIA Tesla V100
    CUDA is pre-installed on the server. You will be given access to your own home directory where you can install your own Anaconda (with Python, packages and frameworks). We currently do not have a particular set of Python or ML framework installed as everyone has different requirements.

    16x Deep Learning Workstations, each equipped with
    GPU: 4 x NVIDIA GTX 2080 Ti, w/ 11GB VRAM Single Blower
    CPU: 1 x 18-Core Intel Xeon W-2295 (36T, 24.75, 3.00 GHz) processor
    Motherboard: Asus C422 SAGE/10G, C422 Chipset, LGA2066, 8DIMMs, up to 256GB DDR4 ECC
    Memory: 256 GB (8x 32GB) DDR4-2933 REG ECC DIMM
    Software installed: CUDA9.2 with matching cuDNN, NVIDIA Docker, Tensorflow, Caffe2, PyTorch, Python, SSH

    6 Deep Learning Workstations, each equipped with
    GPU: 4 x NVIDIA GTX 1080 Ti, w/ 11GB GDDR5X
    CPU: 1 x 8-Core Intel Xeon E5-1680v4 (3.40GHz, 20M Cache, 2400 MHz) processor
    Motherboard: Asus X99-E WS, X99 Chipset, LGA2011-3, 8DIMMs
    Memory: 128 GB (4x 32GB) DDR4-2400 REG ECC DIMM
    Software installed: CUDA9.2 with matching cuDNN, NVIDIA Docker, Tensorflow, Caffe2, PyTorch, Python, SSH

  2. Policy

    The available GPU resources are only opened to the EEE community (i.e. EEE/IEM students, researchers and faculty).

    This GPU policy is put in place to ensure that the limited GPU resources available at MLDA are fairly and optimally utilised. The policy will be reviewed from time to time and updated where necessary.

    Projects that are automatically given an account to use the GPU:

    • MLDA IPPs
    • MLDA FYPs/ISPs
    • MLDA DIPs (1 account per group)
    • MLDA ad-hoc projects (1 account per group)

    For all other users, GPU usage requests will be assessed against a points accumulation system, based on the individual’s contributions made towards MLDA. Only projects parked under MLDA@EEE and workshops and talks/seminars hosted by MLDA@EEE will be counted.

    Users with higher points accumulated will be given priority to use the GPUs. ​


    NTU CITS Acceptable IT Usage Policy
    Users to adhere to the University's IT Usage Policy at all times.
    https://www.ntu.edu.sg/cits/Pages/Acceptable-IT-Usage-Policy.aspx

    Security
    • Ensure that the Operating Systems of the device(s) you use to connect to MLDA GPU servers and workstations are up to date
    • Anti-virus software is installed on your devices and regularly updated. Free Anti-virus software for students and staff is available here: https://citsdownload.ntu.edu.sg/restricted/index.htm

    Backups
    We do NOT create backups of your data on the server/workstations. You are encouraged to backup your work elsewhere.

    Quota
    • Each user will be given a user account under the directory ~/students
    • Contents will be deleted every Semester

  3. To apply for a user account to use the GPU server
    Apply via the registration URL and fill in all the required details:
    https://wis.ntu.edu.sg/webexe88/owa/REGISTER_NTU.REGISTER?EVENT_ID=OA20071122072073​
Login Credentials
If your application is successful, you will be emailed with the following information to log into the server via SSH

​Server IP address IP: 172.28.xxx.xxx ​NTU username login NTU password ​GPU ID
1. Open your SSH client.
2. Type the username provided before the @ symbol (e.g. user1@172.21.xxx.xxx)


Connecting to EEE GPU server remotely via SSH
You will need to install a SSH client to access the GPU server.
  • Mac OS X includes the SSH client Terminal by default.
  • Windows-based Operating Systems do not come with an SSH client by default. We recommend PuTTY. For information, please see: https://www.putty.org/
  • Linux distributions include support for SSH by default as well. Simply start up a terminal. If you are new to command-line interfaces, you may want to start with some of the SSH commands first.
Installing packages in your own user folder
Please install packages within your given user folder and create a local environment. https://www.depts.ttu.edu/hpcc/userguides/application_guides/python.packages.local_installation.php

For packages that require dependencies in C/C++, use conda install cmake
TensorFlow (with GPU support and CUDA dependencies) can be installed in a hassle-free manner (no NVCC or CUDA issues) using Anaconda:
conda install tensorflow-gpu For more information, please visit http://manjusri.ucsc.edu/2017/12/01/anaconda-tensorflow/
​​​​​
Not sure which programme to go for? Use our programme finder
Loading header/footer ...