Fox
Fox has both CPUs and GPUs that you can use, and could be a good choice, since it is new, not too much utilized yet, and easy to get access to. To get access, follow the steps below:
1. Go to https://research.educloud.no/register
2. Select “Apply for access to a project”
3. Click “Next” until you can select which project to join
4. Select “ec69: Advanced AI Course IN5490 ”
5. Wait for approval. It is helpful if you send an email to kaiolae<at>uio.no, so Kai can give you approval faster.
6. Login with ssh <educloud-username>@fox.educloud.no
a. Your Educloud username will likely be “ec-<UiO username>”
See more info about accessing and using Fox here.
If you need access to GPUs on Fox, have a look here.
New: Fox now also has a nice interactive interface available here, for those that have already gotten access with the procedure above.
ML-Nodes
This is the main resource for those of you doing GPU-intensive jobs, such as in Deep Learning. The service is quite easy to get started with, check out their instructions here. The machines are set up to allow you to use typical deep learning frameworks. However if there is any software you are missing on the machines, you can contact the AI-Hub team (itf-ai-support@usit.uio.no) and they are quick to install it for you. You can also install additional Python packages yourself.
- The machine available for our course is called "ml9". You can access it at the address ml9.hpc.uio.no, following the instructions here.
-
It is possible to use these machines through Jupyter which may be user friendly for some students: /tjenester/it/forskning/kompetansehuber/uio-ai-hub-node-project/it-resources/ml-nodes/jupyter_notebooks-.html
-
These machines use a module-system to give you access to software: /tjenester/it/forskning/kompetansehuber/uio-ai-hub-node-project/it-resources/ml-nodes/module-system.html
-
Among other libraries, Tensorflow, Keras, PyTorch and OpenAI Gym are supported through the module system
Saga
Saga is a super computer in Trondheim designed to run heavy duty CPU/GPU jobs. It contains over 300 ndes with each more than 40 cores to run your jobs on. This is an excellent resource for evolutionary experiments to run many jobs in parallel. To get access to saga you will have to fill out the following form: https://www.metacenter.no/user/application/form/notur . Saga uses the SLURM job scheduler. More information on the resource can be found here.
ROBIN-HPC
Check out the Sigma2 NIRD service platform before considering this.
This is a local service at ROBIN that may be useful for those of you running CPU-jobs. You can get access for it by filling out this form. Software: Both pip and conda is installed so you can install python packages (in pip, remember pip install --user <some_package>). For other needs, please contact robin-engineer@ifi.uio.no
ROBIN-workstations
These are local GPU-machines that can be useful if any of you have very specific requirements that mean your experiments cannot run on an external service. There is no queue management on these machines, so the policy is "first comes, first served". If you are a ROBIN-student, you already have access (see how to access the machines here). If you are not a ROBIN-student and need access please contact robin-engineer@ifi.uio.no.
Software: Both pip and conda is installed so you can install python packages (in pip, remember pip install --user <some_package>). For other needs, please contact robin-engineer@ifi.uio.no.