Batchfarm#
The batchfarm of the KTA computer system uses slurm as its workload manager. The offical slurm quickstart guide can be found here.
Submit Queues#
Jobs can be submitted to the batchfarm from the terminal servers. The following queues are available for job submission:
Queue Name |
Purpose |
Max Runtime |
|---|---|---|
|
default queue with all resources |
3h |
|
extended runtime with most powerful nodes |
3d |
|
Very extend runtime with less powerful nodes |
7d |
|
for running jobs which require gpu |
1d |
|
short test runs and debugging |
10min |
Users are encouraged to use the test queue for quick checks and debugging to avoid unnecessary resource consumption in the main queues. More information about the different queues can be gained by running sinfo or scontrol show partition <NAME>.
Compute nodes#
The compute nodes cannot be accessed by users and used interactively. All computational tasks have to be submitted via slurm. The following compute nodes are available:
Host name |
# Cores/Threads |
CPUs |
GPUs |
RAM (GB) |
|---|---|---|---|---|
128/128 |
AMD EPYC 7713 |
- |
512 |
|
128/128 |
AMD EPYC 7713 |
- |
512 |
|
128/128 |
AMD EPYC 7713 |
- |
512 |
|
128/128 |
AMD EPYC 7713 |
- |
512 |
|
128/128 |
AMD EPYC 7713 |
- |
512 |
|
128/128 |
AMD EPYC 7713 |
3x NVIDIA L40 |
512 |
|
40/40 |
Xeon Gold 6148 |
4x NVIDIA GTX 1080 Ti |
92 |
|
40/40 |
Xeon Gold 6148 |
4x NVIDIA GTX 1080 Ti |
92 |
|
24/24 |
Xeon E5-2690 |
4x NVIDIA GTX 980 |
256 |
|
40/40 |
Xeon Gold 6148 |
- |
187 |
|
40/40 |
Xeon Gold 6148 |
- |
187 |
|
40/40 |
Xeon Gold 6148 |
- |
187 |