Auto-generated from results/<timestamp>/ snapshots on every push to main.
🕐 Latest run
2026-07-19T10:16:25Z
🖥️ Largest fleet tested
10000nodes / 80000GPUs
⚡ Best sched. throughput
142.8pods/s
🧪 Experiment runs
| Run | Timestamp | 🖥️ Nodes | 🎮 GPUs | 📦 Pods | ⚡ Sched. throughput (pods/s) | ⏱️ Sched. e2e p99 (s) | 🗄️ etcd DB (MB) |
|---|
| 20260719-121534 | 2026-07-19T10:16:25Z | 10000 | 80000 | 74381 | 13.7 | 79.848 | 578.2 |
| 20260719-103101 | 2026-07-19T08:34:01Z | 1000 | 8000 | 10500 | 13.5 | 69.675 | — |
| 20260719-102054 | 2026-07-19T08:21:02Z | 1000 | 8000 | 21546 | 142.8 | 0.530 | 550.9 |
| 20260719-090923 | 2026-07-19T07:09:25Z | 1000 | 8000 | 8000 | 28.2 | 0.154 | 116.6 |
| 20260719-090446 | 2026-07-19T07:04:47Z | 1000 | 8000 | 1000 | — | — | 14.5 |
| 20260719-085932 | 2026-07-19T06:59:32Z | 1 | 8 | 1 | 0.0 | 0.010 | 0.5 |
🏁 ClusterLoader2 runs
| Run | p50 | p90 | p99 | max |
|---|
| cl2-20260719-105723 | 188.4 | 192.8 | 192.8 | 192.8 |
SchedulingThroughput (pods/s), official upstream framework.