GPGPU computing servers

These servers are intended for GPU computing (GPGPU), aimed at compute-intensive tasks, machine learning, data processing and scientific simulation that require acceleration by graphics hardware.

Any researcher at the center can request access to these servers. Access is granted after application and validation.

• Servers in the HPC computing cluster: HPC computing cluster
• Servers at CESGA: Request access

Access to these servers is restricted to a specific group, a particular project, or is more tightly controlled for resource management and planning reasons.

It is essential to check the updated information in Xici at the time of requesting the service, where the particular conditions of each server are detailed (access criteria, priorities, usage conditions, etc.).

⚠️ The servers ctgpgpu15, ctgpgpu16, ctgpgpu17 and ctgpgpu18 have a temporary installation and assignments, and their configuration and accesses could be changed in the future.

Not all servers are available at all times for any use. To access the servers, you must request it in advance through the incident form. Users who do not have access permission will receive an incorrect password message.

To connect to the servers, you must do so via SSH. The names and IP addresses of the servers are as follows:

Connection is only available from the center's network. To connect from other locations or from the RAI network it is necessary to use the VPN or the SSH gateway.

On servers where a Slurm queue manager is present, its use is mandatory to submit jobs and thus avoid conflicts between processes, since two jobs must not be run at the same time. Users are limited to using two cores and 1GB of RAM outside Slurm.

To submit a job to the queue use the “sbatch” command with a Slurm script or directly srun:

srun cuda_program program_arguments

The srun process waits for the job to run before returning control to the user. If you do not want to wait, session managers such as screen can be used, allowing you to leave the job running and disconnect the session without worry and recover the console output later.

Alternatively, you can use nohup and put the job in the background with &. In this case the output is saved in the nohup.out file:

nohup srun cuda_program program_arguments &

To see the queue status use the squeue command. The command shows output similar to this:

JOBID PARTITION     NAME     USER  ST       TIME  NODES NODELIST(REASON)
9  servidore ca_water pablo.qu    PD       0:00      1 (Resources)
10 servidore ca_water pablo.qu    PD       0:00      1 (Priority)
11 servidore ca_water pablo.qu    PD       0:00      1 (Priority)
12 servidore ca_water pablo.qu    PD       0:00      1 (Priority)
13 servidore ca_water pablo.qu    PD       0:00      1 (Priority)
14 servidore ca_water pablo.qu    PD       0:00      1 (Priority)
 8 servidore ca_water pablo.qu     R       0:11      1 ctgpgpu2

You can also obtain an interactive view, updated every second, with the smap command:

smap -i 1

On servers that use gpuctl, it is necessary to use the gpu command to request GPUs and be able to run jobs on them.

The gpu command does not include a job queue or parallelism control. It can be used interactively or from scripts, but to queue multiple jobs or run them in parallel you should use external tools, such as task-spooler (command tsp).

There are two main workflows:

This method is best if you only want to run a single command. gpu exec acts as a wrapper: it takes care of claiming the GPU, preparing the necessary environment for the command and releasing it when finished.

gpu exec python ./train.py

This is the recommended option for simple runs, as it avoids having to manually manage the reservation and release of the GPU.

This method is best if you want to keep a reservation active across several commands, either in an interactive session or in a workflow with several consecutive runs.

gpu claim
gpu exec python ./train.py
gpu release

Or, alternatively:

gpu claim
eval "$(gpu env --shell)"
python ./train.py
gpu release

If a reservation already exists, gpu exec reuses it and automatically prepares the appropriate environment for the command.

It is important to remember that after creating or modifying a reservation, you must run eval "$(gpu env --shell)" again if commands will be launched directly from the shell. Alternatively, you can use gpu exec, which prepares the environment on each execution.

Reservations are kept as long as real compute activity is detected on the GPU.

If a GPU remains reserved without activity for a prolonged time, the reservation may be lost automatically due to inactivity. Therefore, if after that you try to run a job assuming the reservation is still active, the job may fail.

The reservation is not maintained simply by having a terminal open, but by real activity on the GPU.

The inactivity expiration times are configured as follows:

• Monday to Friday, from 10:00 to 18:00 → 1 hour
• Other times → 2 hours
• If the reservation appears abandoned (no user process on the system nor any shell) → 10 minutes

If you are working with a manual reservation and want to check if it is still active, you can use:

gpu mine

Or, if you want to see general information about all GPUs, the time when real usage was last observed for each, and the current configured expiration time, you can run:

gpu status

This is especially recommended if time has passed since the last run or if the GPU may have been unused for a long period.

Although gpu does not implement a job queue as such, there is an implicit waiting system for GPU reservation.

When a command that requires reserving a GPU is executed and none are available, the command does not fail but instead remains blocked waiting for a GPU to become free. The user enters an internal queue managed by gpuctl and the execution will continue automatically as soon as the reservation can be satisfied.

This behavior applies to both manual and automatic reservations.

For example:

gpu exec python ./train.py 

If no GPUs are available, this command will wait until one is released.

Similarly:

gpu claim 

will also remain blocked until it can obtain a GPU.

By default, gpuctl automatically assigns an available GPU. However, it is possible to request a specific GPU by its index.

Both gpu claim and gpu exec allow the –gpu-index parameter for this purpose:

gpu claim --gpu-index 0 
gpu exec --gpu-index 1 python ./train.py 

In this case, the command will attempt to reserve the specified GPU. If that GPU is not available, the command will wait in queue until that particular GPU is released.

This behavior allows fixing runs to specific GPUs, but may increase waiting time if the chosen GPU is in high demand.

In addition to the main reservation, which is made with priority in the guaranteed queue, it is possible to reserve a second GPU in the burst_preemptible queue on some servers (check the reservation policy message when entering the server; if applicable it will refer to a second opportunistic GPU).

This second GPU can be used whenever it is free, but it is not guaranteed: it can be lost even in the middle of a job if another user needs that GPU for their primary reservation.

This allows taking advantage of additional free capacity when available, but jobs that depend on this second GPU must be prepared to tolerate its loss.

To request this second GPU, it is enough to run gpu claim a second time:

gpu claim
gpu claim

You can also explicitly specify the number of GPUs:

gpu claim --numgpus 2

In this case, the command will wait until it can reserve the requested number of GPUs.

Note that these two forms are not exactly equivalent:

• with gpu claim followed by another gpu claim, the first reservation is obtained earlier and the second is requested later;
• with gpu claim ---numgpus 2, the request is made jointly and the command will wait until both GPUs can be reserved.

It is important to remember that after making any reservation you must run eval "$(gpu env --shell)" again or perform executions with gpu exec.

If the second GPU is lost later for being preemptible, the jobs using it could fail or be interrupted.

Therefore:

• the first GPU is the priority, guaranteed reservation;
• the second GPU is opportunistic and will only be available while it is not needed for another priority reservation;
• if any GPU cannot be reserved, the command will remain waiting and the user will enter the corresponding queue;
• when reserving two GPUs at once, the command will wait until both can be reserved at the same time.

Since gpu does not include a job queue or parallelism control, a practical option to queue executions is to use task-spooler, with the tsp command, combined with gpu.

gpu claim
tsp gpu exec python ./train1.py
tsp gpu exec python ./train2.py
tsp gpu exec python ./train3.py
tsp gpu release

This method allows reusing the same reservation for several consecutive jobs queued in tsp.

If you want to run several jobs at the same time, you can increase the number of slots of tsp:

gpu claim
tsp -S 2
tsp gpu exec python ./train1.py
tsp gpu exec python ./train2.py
tsp gpu exec python ./train3.py
tsp gpu exec python ./train4.py
tsp -w
tsp -S 1
gpu release

In this case, be careful not to run gpu release before all jobs have finished. First make sure the queue has finished, and only then release the GPU.

Attention: running several jobs in parallel does not imply reserving multiple GPUs. If only one GPU is reserved, all processes will share that same GPU and compete for its resources, such as memory and compute time.

Also remember that the reservation is only maintained while real compute activity on the GPU is detected. If the jobs go too long without using it, the reservation may automatically expire due to inactivity.

Using tmux allows keeping terminal sessions, disconnecting from them and reattaching later, thus maintaining an interactive session without keeping the connection open. Disconnecting also does not send HUP to processes.

To start a tmux session or reconnect to it if it already exists, it is recommended to use a fixed name:

tmux new -A -s main

This creates the main session if it does not exist, or reconnects to it if it already exists.

Once inside tmux, to detach while keeping the session active press Control+b, release, and then press d.

At this point you can disconnect from the machine without interrupting processes. To return to the same session, just run the previous command again.

If you want to terminate the session completely from outside:

tmux kill-session -t main

gpuctl allows partitioning a GPU into several MIG instances to isolate resources and run smaller workloads in parallel on a single physical GPU in a more controlled way.

This functionality is only available on compatible GPUs; otherwise it cannot be used. So far, only NVIDIA H200 GPUs have this function enabled.

To reserve a GPU and enable it in MIG mode:

gpu claim --mig half
gpu claim --mig third
gpu claim --mig quarter

This will split the GPU into two, three or four partitions, choosing the most appropriate MIG profiles according to the compatible GPU. Note that this parameter --mig can only be used when exactly one GPU is claimed.

Once the MIG reservation is created, gpu exec and gpu env automatically prepare CUDA_VISIBLE_DEVICES with the corresponding MIG UUIDs.

gpuctl allows configuring the sending of email notifications related to the state of reservations and GPU usage.

gpu notify

gpu notify --mode off
gpu notify --mode email
gpu notify --mode telegram
gpu notify --mode both

When notifications are enabled, emails or Telegram messages are sent in the following cases:

• when a GPU is automatically released for remaining unused;
• when the enforcer kills a process for using a GPU without having an active reservation or for losing an opportunistic GPU;
• when a reservation request is queued;
• and later when that reservation is granted.

Notifications are not sent in case a reservation is made and the reservation is granted immediately without needing to wait, or if the GPU is released manually with release or at the end of an exec that made an automatic reservation.

gpu notify --email your.email@domain

If a custom email is not configured, the address obtained from LDAP will be used by default. To revert to the default email:

gpu notify --clear-email

After running the following command:

gpu notify --pair-telegram

Follow the instructions to complete the pairing process. To remove the pairing:

gpu notify --unpair-telegram

The user can send custom notifications at various points in their workflow using:

gpu notify --send "message content"

This can be useful to receive personal alerts at the start or end of an experiment, after completing a processing phase, or at any other relevant point in the workflow.

The preferred notification method configured will be used, or email if neither is configured.

To consult a summary of the recent historical GPU usage by process, you can use:

 gpu report 

This shows the recorded PIDs, number of samples available and aggregated metrics of GPU, CPU and memory usage.

If you want to see data for a specific process, with ASCII graphs and the timeline of stored samples, you can use:

 gpu report --pid PID 

If there are many samples, the output may be summarized. To see the full series:

 gpu report --pid PID --all-samples 

• To run a single command, the simplest is usually gpu exec.
• To keep a reservation active across several commands, it is more appropriate to use gpu claim + gpu exec + gpu release.
• If you will work directly from the shell after making a manual reservation, run eval "$(gpu env --shell)" or prefix commands with gpu exec.
• To leave sessions active and recover them later, tmux is usually the most convenient option.
• To queue multiple jobs, tsp is a practical alternative.
• If you use tsp with parallelism, remember that this does not reserve additional GPUs: processes will share the GPU(s) visible at that moment.
• If a reservation remains too long without real compute activity on the GPU, it may be lost automatically due to inactivity.