Create Job

Job Definition

Show advanced fields

Core info

Project

Select a project

Name

Description

Describe what your experiment does (optional)

Docker Container

Image

Command

When left empty, the command specified in the Dockerfile from the image will run

User

When left empty, the user specified in the Dockerfile from the image will be used.

Environment

Required Resources

Cluster ID

Leave empty to run your job on any compatible cluster

GPU

GPUs

GPU Model

The required GPU model for this job. A partial name is allowed: '1080' will match a 'GeForce GTX 1080 Ti' GPU

GPU Memory

The minimum amount of GPU memory required (per GPU). If you request 5 GB, and get assigned GPUs with 8 GB, you will have the full 8GB available for each GPU.

CPU

CPU cores

CPU Memory

Minimum CUDA Version

The minimum CUDA driver version required for this job.

Slave ID

The name of the slave this job should run on.

Instance ID

The name of the instance this job should run on.

Input and output

Storage

Mount tmpfs

tmpfs size

The amount of extra CPU memory that will be allocated to the tmpfs
Total CPU memory used by this job: GB + GB = NaN GB

tmpfs path

The directory inside the container to which the tmpfs should be mounted.

Port Mappings

Scheduling

Interactive

Interactive jobs will either run immediately, or fail directly. They will never be QUEUED for a long time.

Restartable

Restartable jobs can be halted and later restarted by GPULab. Before stopping, GPULab will send a SIGUSR1 signal to the job. The job then needs to stop with exit status 123 within 15 seconds. In exchange for this flexibility, GPULab can start the jobs sooner than it otherwise would.

Max Duration

The maximum duration of this job. GPULab will always stop your job after this time. If the job has been restarted once or more, the total duration of all the job runs is used.
Format: a number followed by a unit (ex: 5 minutes, 3 hour, 2 days, 1 week, ...)

Not Before

GPULab will not start the job before the specified date. It will remained QUEUED.
Format: an RFC3339 date.

Not After

GPULab will set the job to FAILED if the job is still QUEUED at the specified date. (Does not affect an already running job, only prevents start after the date.)
Format: an RFC3339 date.

minDuration

GPULab might halt this job after this duration. It will not halt the job before this duration is reached. Setting this allows GPULab to schedule this job earlier than it can otherwise, but there is a chance your job will be stopped after this time. (If it is restartable, it will however restart later.) Tip: If your job can be halted, set this to the minimum time your job needs to do useful work. For example, If your job needs 10 minutes to prepare the dataset before starting to work, set this to at least 30 minutes.
Format: a number followed by a unit (ex: 5 minutes, 3 hour, 2 days, 1 week, ...)

reservation

Please select 0, 1 or more Reservations

Choose the Reservation to use for this job.

Limit Maximum Simultaneous Jobs

Manually control the parallelism of your jobs

Bucket Name

Unique ID shared between jobs of which you want to limit the number of them running in parallel

Bucket Max

The maximum number of jobs that can run inside the bucket with the same name

Email on queue

An email address that will be informed when the job is queued.

Email on run

An email address that will be informed when the job has started.

Email on end

An email address that will be informed when the job is finished.

Extra SSH Key

An extra SSH key to add, in addition to the key of the user creating the job.

Start the job

Resulting JSON:

"root"

{}