Skip to content
GitLab

1.a. Supercomputers · Changes

Page history
Update 1.a. Supercomputers authored by Okke van Eck's avatar Okke van Eck
Hide whitespace changes
Inline Side-by-side
1.a.-Supercomputers.md
View page @ 665b66cc
...@@ -14,7 +14,7 @@ The Lumi supercomputer is divided in two partitions: [Lumi-C](https://docs.lumi- ...@@ -14,7 +14,7 @@ The Lumi supercomputer is divided in two partitions: [Lumi-C](https://docs.lumi-
The internal hardware structure of these partitions is different, and thus a comparison between the two is mostly of no use. The internal hardware structure of these partitions is different, and thus a comparison between the two is mostly of no use.
In other words, the CPU version of your code should be executed on Lumi-G as well when comparing to the GPU version. In other words, the CPU version of your code should be executed on Lumi-G as well when comparing to the GPU version.
One Lumi-G node uses 4 AMD MI250X GPUs in combination with a single 64-cores AMD EPYC 7A53 "Trento" CPU. The CPU offers 8 processes, of which 1 is dedicated to running the OS. Each node thus offers 7 processes to be executed in parallel. One Lumi-G node uses 4 AMD MI250X GPUs in combination with a single 64-cores AMD EPYC 7A53 "Trento" CPU. The CPU offers 8 threads, of which 1 is dedicated to running the OS. Each node thus offers 7 processes to be executed in parallel.
### Launching jobs ### Launching jobs
Lumi uses the [Slurm Workload Manager](https://slurm.schedmd.com/documentation.html), and thus jobs can be launched using a simple `.batch` script. Running interactive nodes is a bit of a hassle since Lumi does not support SSHing into compute nodes. Luckily there is a workaround. Lumi uses the [Slurm Workload Manager](https://slurm.schedmd.com/documentation.html), and thus jobs can be launched using a simple `.batch` script. Running interactive nodes is a bit of a hassle since Lumi does not support SSHing into compute nodes. Luckily there is a workaround.
......