Commit 21332346 authored by Josef Hrabal's avatar Josef Hrabal

Merge branch 'dgx' into 'master'

Dgx

See merge request !250
parents 37559e72 b8a97cf7
Pipeline #6558 passed with stages
in 1 minute and 39 seconds
......@@ -5,6 +5,13 @@
!!! warning
GPUs are single-user devices. Memories of each GPU are not purged between job runs. Furthermore, they can be read (but not written) by any user at any time. Consider the confidentiality of your running jobs.
## Shell Access
## How to Access
The DGX-2 machine can be accessed by SSH protocol via login nodes at the address `loginX.salomon.it4i.cz`.
!!! info
Should you need access to the DGX-2 machine, request it at support@it4i.cz.
The DGX-2 machine can be accessed through the scheduler from `loginX.salomon.it4i.cz` nodes.
The NVIDIA DGX-2 has its own instance of the scheduler, it can be accessed after loading the `DGX-2` module. See [Resource Allocation and Job Execution][1].
[1]: job_execution.md
\ No newline at end of file
......@@ -2,6 +2,9 @@
To run a job, computational resources for this particular job must be allocated.
!!! info
You can access the DGX PBS scheduler by loadnig the "DGX-2" module.
## Resources Allocation Policy
The resources are allocated to the job in a fair-share fashion, subject to constraints set by the queue. The queue provides prioritized and exclusive access to computational resources.
......@@ -27,9 +30,6 @@ When allocating computational resources for the job, specify:
!!! note
Right now, the DGX-2 is divided into 16 computational nodes. Every node contains 6 CPUs (3 physical cores + 3 HT cores) and 1 GPU.
!!! info
You can access the DGX PBS scheduler by loadnig the "DGX-2" module.
Submit the job using the `qsub` command:
**Example for 1 GPU**
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment