If you cannot use your personal GPU, there are 2 options for training: 

• Option 1: Use the classroom computers with GPUs when they are not in use during the day or remotely during the night and weekends.

• Option 2: We provide you access to the CBIA computational server beta.fi.muni.cz. (Accessible via SSH and SFTP using the faculty login.)

Option 1

A few computers with GPUs in the computer classroom (sirene01–sirene05 in room B311) are available via remote access on weekends and weekdays from the end of classes to an hour before classes start (approximately from 10 pm to 6 am). More information on this can be found here: https://www.fi.muni.cz/tech/win/classrooms.html.en#rdesktop  . 

Option 2 

We provide you access to the CBIA computational server beta.fi.muni.cz. Until the deadline of 12th May, there is one GPU (Nvidia A1OO, 80 GB) dedicated only to you. This GPU was split into 4 partitions of 20 GB of memory.

These partitions are named GPU_A, GPU_B, GPU_C and GPU_D. 

Only one user may work with one GPU partition at a time to ensure that there are no conflicts.

We prepared a simple reservation system for GPU partitions. For each partition, there are two twelve-hour time slots per day. To reserve a slot, write your name in the proper cell of the shared document. There are 42 students in the course, so on average, everyone should be able to reserve 8 slots. If free slots are available, it is allowed to reserve more, but only if it does not prevent other students from working on their own projects.

https://docs.google.com/spreadsheets/d/1q2Ryi8dv5mAGGt9RbKq48_Z-tulkNiXgr21fAUQlMKA/edit?usp=sharing 

You can log into the server using your faculty login via ssh or VNC. Before you start the GPU computation, make sure to choose only the GPU partition you reserved, e.g., by running the command:

    export CUDA_VISIBLE_DEVICES=${GPU_C}

After you finish the computation, make sure to stop all of your processes that use the GPU.

More technical details can be found on the CBIA support page.

https://cbia.fi.muni.cz/services/support.html

Lastly, we recommend starting your project early on, when the server is less occupied. Based on previous years' experience, the server will probably be fully occupied a few days before the deadline.

Do not hesitate to ask us any questions by email or during the seminars.