User Tools

Site Tools


wiki:prototype

ARM-based clusters

Different cluster states:

  • Dismantled: The cluster is no longer available. In case of need of this platform, please contact us and we will study how to provide the required node configuration.
  • Deploying: The cluster is in deployment, meaning that we are setting up everything. In other words, the cluster is not available but it will be available soon.
  • Not avail.: The cluster is not available for some reason and no date for being available again can be provided.
  • Available: The cluster is currently available to be used and, therefore, is accessible.
Basenode Cluster login node Status # of nodes Architecture
CARMA carma Dismantled 10 ARMv7-a
Arndale arndale Dismantled 3 ARMv7-a
ODROID-XU odroid Dismantled 24 ARMv7-a
Jetson TK1 jetson-1 Dismantled 8 ARMv7-a
ODROID-XU3 octodroid-1 Available 13 ARMv7-a
JUNO r1 juno Not avail. 2 ARMv8-a
Jetson TX1 (2) jetsontx-1 Available 16 ARMv8-a
Merlin (1) merlin-4 Available 4 ARMv8-a
ThunderX (1) HCA Available 4 ARMv8-a
Falkor HCA Available 4 ARMv8-a

For information about the Mont-Blanc prototype, please refer to this other wiki.

(1) Merlin and ThunderX share the same login node, but some, libraries should remain specific for each cluster. In this matter, some modules are split in different categories so you can choose the version for an specific platform. Where this case applies, just use the module of the cluster you plan to run your binary at.

(2) Jetson TX1 cluster shares some applications with Merlin and ThunderX. In this matter, some modules are split in different categories so you can choose the version for an specific platform.

Contacts

The team of system administrators of Mont-Blanc is available to assist you in more detailed configuration. Do not hesitate to contact us at hca.sysadmin@bsc.es.

Using OpenMPI

Due to a bug in the vader module, you must execute your jobs with the flag --mca btl ^vader

More details here

Compiler optimization flags

Arm HPC Compiler

armclang -mcpu=native -O3 -ffast-math
 
# Warning: For Cavium ThunderX2, -mcpu=native and -mcpu=thunderx2t99 yield different results
# We suggest using -mcpu=thunderx2t99
armclang -mcpu=thunderx2t99 -O3 -ffast-math

If you are using fortran, consider adding -fstack-arrays

Please refer to the Porting and Tuning guides for various packages page of the Arm developer portal for more information.

GCC

gcc -march=native -mcpu=native -O3 -ffast-math -ffp-contract=fast
wiki/prototype.txt · Last modified: 2019/01/22 21:45 by kpeiro