A production AI HPC system using fully open Ethernet networking achieves top-100 performance while documenting typical single-tenant LLM workload patterns of many small jobs consuming little time and few large jobs dominating GPU hours.
Congestion control for large-scale RDMA deployments
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.DC 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
SAKURAONE: An Open Ethernet-Based AI HPC System and Its Observed Workload Dynamics in a Single-Tenant LLM Development Environment
A production AI HPC system using fully open Ethernet networking achieves top-100 performance while documenting typical single-tenant LLM workload patterns of many small jobs consuming little time and few large jobs dominating GPU hours.