OPA
E500306
OPA is a high-performance computing interconnect architecture developed by Intel to provide low-latency, high-bandwidth communication in large-scale clusters and supercomputers.
Statements (46)
| Predicate | Object |
|---|---|
| instanceOf |
Intel technology
ⓘ
high-performance computing interconnect architecture ⓘ |
| category |
cluster interconnect
ⓘ
supercomputer interconnect ⓘ |
| communicationModel | switched fabric ⓘ |
| competesWith |
Ethernet-based HPC interconnects
ⓘ
InfiniBand NERFINISHED ⓘ |
| connects |
compute nodes in clusters
ⓘ
storage systems in HPC environments ⓘ |
| deploymentScale |
tens of thousands of nodes
ⓘ
thousands of nodes ⓘ |
| developer | Intel NERFINISHED ⓘ |
| domain | high-performance computing ⓘ |
| feature |
collective offload capabilities
ⓘ
end-to-end congestion control ⓘ fabric management tools ⓘ hardware-based transport offload ⓘ high bandwidth ⓘ low latency ⓘ quality of service mechanisms ⓘ scalability ⓘ |
| fullName | Omni-Path Architecture NERFINISHED ⓘ |
| goal |
improve scalability of parallel applications
ⓘ
reduce communication bottlenecks in HPC ⓘ |
| introducedBy | Intel Data Center Group NERFINISHED ⓘ |
| layer | system interconnect ⓘ |
| manufacturer | Intel NERFINISHED ⓘ |
| optimizedFor |
large-scale MPI jobs
ⓘ
parallel computing workloads ⓘ |
| provides |
fabric management software stack
ⓘ
host channel adapters ⓘ switches ⓘ |
| purpose |
high-bandwidth communication
ⓘ
low-latency communication ⓘ |
| supports |
HPC applications
ⓘ
RDMA-like communication semantics ⓘ message passing interface workloads ⓘ |
| targetHardware |
HPC clusters
ⓘ
supercomputing systems ⓘ |
| transportType | packet-switched interconnect ⓘ |
| typicalTopology |
dragonfly
ⓘ
fat-tree ⓘ |
| useCase |
large-scale clusters
ⓘ
supercomputers ⓘ |
| usedIn |
HPC data centers
ⓘ
technical computing environments ⓘ |
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.