CPU Compute Lab
The CPU Compute Lab provides access to latest generation CPU technologies, to enable code porting and optimisation by UK researchers in preparation for future large production systems.
Current systems include
-
Direct ssh to the gi001 node
Via the dine2 Slurm partition
Bergamo (2023)
Via the cosma5 Slurm partition
Genoa (2023)
Via the cosma8-shm3 Slurm partition
IceLake (2022)
Via the cosma8-ska Slurm partition
Milan (2021)
Via the cosma8-milan Slurm partition
Rome (2019)
Via the cosma8-rome Slurm partition
CascadeLake (2019)
Via the cosma7-shm2 Slurm partition
More on Sapphire Rapids
Node access
Our Sapphire Rapids are configured to hold Ponte Vecchio GPUs (PVCs). Therefore, they are accessible as part of our Intel GPU nodes.
(*) The nodes are given out FCFS, i.e. please check prior to any benchmarking that nobody else is currently using the node.
Specification
CPU name: Intel(R) Xeon(R) Platinum 8480+
Sockets: 2
Cores per socket: 56
Threads per core: 2
Environment
As this is an Intel node while Cosma’s login nodes are AMD, we recommend to compile exclusively on the node:
module load intel_comp
module load compiler-rt tbb compiler mpi
The module oneAPI should work as well.
Performance
To assess the node’s performance, we recommend to run
module load likwid
likwid-bench
Type in -a or -h to see an overview of the available metrics. Stream Triad and peak are the ones we usually recommend to look at first.