CUDA, OpenCL, Python | Wydział Elektroniki, Telekomunikacji i Informatyki Politechniki Gdańskiej

Basic infromation

All computers should have python-virtualenv installed. If you need other version than the one provided with the system please use it to install it manually,

All CUDA systems are docker/nvidia docker enabled. Docker access is however limited and granted individually!

GPU Reservation

Starting 2022-01-10 GPU reservation is mandatory for most of the hosts. Currently mandatory reservation is required for hosts:

• apl11.maas
• apl13.maas
• apl15.maas
• apl16.maas
• des19.kask
• des20.kask
• des21.kask
• sanna.kask
• vinnana.kask

It is recommended to make reservations for hosts:

• apl12.maas

The vedana.kask host is currently not accessible for general usage. You do not have to reserve apl12 but users with reservations have priority! If you will schedule calculations without making a reservation you will be notified via e-mail. Users not making or respecting reservations will be banned from the GPU servers.

Reservation rules

1. Students
   1. you can reserve up to half of the GPUs in a given host
   2. only GPUs from one host can be reserved at the same time
   3. the reservation can be done for up to 3 days
2. Faculty members
   1. you can reserve any number of the GPUs
   2. you can reserve any number of hosts
   3. the reservation can be done for up to 7 days

In both cases after finishing calculations please close all unnecessary sessions, screen, tmuxes, etc.

TensorHive system

To make a reservation please go to the following site:

https://kask.eti.pg.gda.pl/tensorhive/

You have to register providing the user name equal to the one in the laboratory. Before hitting the register button you have to append the key displayed there to your .ssh/authorized_keys file on any of the KASK servers, preferably on kask.eti.pg.gda.pl. Without that the registration will fail as this serves as a check that you have a valid KASK account.

Computers in room 527

des01.kask - des18.kask

• RTX 4070 Ti 6GB
• CUDA: 11.8
• OpenCL: 1.2 INTEL + 3.0 CUDA
• Python 3.10
• tensorflow (2.14.0) - WITH GPU SUPPORT

des19.kask - des20.kask

• GTX 1070 8GB
• CUDA: 11.7
• OpenCL: 1x OpenCL 3.0 CUDA
• Python 2.7 (deprecated) and Python 3.8
• tensorflow (2.4.1) - WITH GPU SUPPORT

des21.kask

• GTX 1660 6GB
• CUDA: 11.7
• OpenCL: 1x OpenCL 3.0 CUDA
• Python 2.7 (deprecated) and Python 3.8
• tensorflow (2.4.1) - WITH GPU SUPPORT

Servers

If not stated otherwise the servers have the following software versions installed:

• CUDA: 11.8
• OpenCL: OpenCL 3.0 CUDA
• Python 3.10
• tensorflow 2.13.0

If you need other versions please use python virtual env or local python packages!

If local conditions allow the local storage is made available for users. The local HDD space is available under /local_storage and SSD under /ssd_local. There is common directory writable by all users that can be used to store temporary data.

The /ssd_local and /local_storage directories should be temporary only! Files belonging to users that did not login for extended period of time can be deleted without any warning!

apl11.maas

• RTX A4500 20GB
• A100 80GB
• local_storage - 3 TB (shared with the system!)

apl12.maas

• RTX 2080 Ti 11GB
• local_storage - 1.7 TB

apl13.maas

• 2x Quadro RTX 8000 48GB
• 1x Quadro RTX 5000 16GB
• local_storage - 11 TB
• ssd_local - 1.7 TB

apl15.maas

• RTX 2080 8GB
• local_storage - 1.3 TB

apl16.maas

• 2x RTX 2080 8GB
• local_storage - 1.3 TB

sanna.kask(172.20.97.101)

• 8x Quadro RTX 6000 24GB
• local_storage - 3.6 TB

vinnana.kask (172.20.97.102)

• 4x Quadro RTX 5000 16GB
• local_storage - 3.6 TB

vedana.kask (172.20.97.103)

• 4x Quadro RTX 5000 16GB
• CUDA: 11.7
• OpenCL: OpenCL 3.0 CUDA
• Python 2.7 (deprecated) and Python 3.8
• tensorflow (2.14.0) - WITH GPU SUPPORT