Validated Hardware and Software Components
The following table summarizes the software versions tested and validated by role for this JVD.
Table 13: Validated Hardware and Software
| Device / Software | Role | Version / Release |
|---|---|---|
| QFX devices | Frontend leaf and spine nodes | 25.2X100-D20.4-EVO |
| AMD MI300X server | Inference GPU server |
Ubuntu 22.04.5 LTS 6.8.0-111-generic |
| HPE Juniper Apstra Data Center Director | Fabric automation and operations | 6.1 |
| SGLang | Inference serving framework | 0.4.5 |
| SGLang Router | Request routing across local GPU-backed workers | 0.1.4 |
| Envoy Proxy Service | Load balancing across MI300X inference servers | 1.35.3 |
| NVIDIA GenAI-Perf | Inference benchmark load generator | Nvidia GenAI-Perf 0.0.11 |
| Models | Inference workloads used to validate performance across different model sizes, concurrency levels, latency profiles, and frontend fabric traffic patterns. |
Llama 3.1 8B Llama 3.3 70B Qwen 2.5 72B |
Note: The Juniper products and software versions
listed above pertain to the latest validated configuration for this
JVD. Different hardware models and software versions may be tested
and added to future design recommendations.