System
Computational Resources
Tempest is an HPC cluster consisting of EPYC X86 CPU and Nvidia GPU nodes. In total, Tempest currently has 12,264 logical CPU cores, 55.9TB of ECC memory, 26 Nvidia A40 GPUs, 14 Nvidia A100 GPUs, and 12 Nvidia H100 GPUs. For storage, Tempest currently has 600TB of all-flash high-speed storage. Tempest components are interconnected by 100/50/25Gb Ethernet with RDMA.
For more information, see the detailed description of computational resources
Cluster Partitions
Tempest currently has the following partitions:
Partition
|
Descritpion
|
---|---|
priority
|
The primary priority-access partition
|
unsafe
|
Preemptible partition with access to all unused CPU node resources
|
test
|
Partition for small test jobs
|
gpupriority
|
The primary priority-access GPU partition
|
gpuunsafe
|
Preemptible partition with access to all unused GPU node resources
|
gputest
|
Partition for small GPU test jobs
|
legacy
|
Fairshare partition with older hardware
|
nextgen
|
Fairshare partition with the newest hardware
|
nextgen-long
|
Same as nextgen, but with a longer max runtime
|
nextgen-gpu
|
Fairshare partition with the newest GPU hardware
|
nextgen-gpu-long
|
Same as nextgen-gpu, but with a longer max runtime
|
For a more detailed description of the partitions, see the partition information.
For guidance on which partition to use, see the partition use case documentation.