@@ -12,11 +12,11 @@ On Turing node by default gpu partition offers 384 GB of usable memory. The memo
...
@@ -12,11 +12,11 @@ On Turing node by default gpu partition offers 384 GB of usable memory. The memo
- 8 GB per reserved CPU core if hyperthreading is deactivated (Slurm option `--hint=nomultithread`).
- 8 GB per reserved CPU core if hyperthreading is deactivated (Slurm option `--hint=nomultithread`).
The default gpu partition is composed of 4 GPUs and 48 CPU cores: you can reserve for instance 1/4 of the node memory per GPU by reserving 12 CPU cores (i.e. 1/4 of 48 CPU cores) per GPU:
The default gpu partition is composed of 4 GPUs and 48 CPU cores: you can reserve for instance 1/4 of the node memory per GPU by reserving 12 CPU cores (i.e. 1/4 of 48 CPU cores) per GPU. However, **it is suggested to choose a slightly smaller such as 10** value as the server does not allow the full utilisation of the GPUs.
--cpus-per-task=12 # reserves 1/4 of the node memory per GPU (default gpu partition)
--cpus-per-task=10 # reserves ~1/4 of the node memory per GPU (default gpu partition)
In this way, you have access to 96 GB of memory per GPU if hyperthreading is deactivated (if not, half of that memory).
In this way, you have access to 80 GB of memory per GPU if hyperthreading is deactivated (if not, half of that memory).