SkyPilot rack-aware scheduling adds usable GPU capacity
What happened
On October 29, Hacker News' front page reported that SkyPilot raises usable GPU capacity on Grace Blackwell NVL72 through topology-aware scheduling, in-place recovery from failures and automatic defragmentation of shared clusters. The report ties this to the need for a new scheduler: SkyPilot keeps multi-node jobs local to a rack and, with failure recovery and cluster defragmentation, gets 27% more usable GPUs from the same rack. The disclosed progress covers these scheduling and cluster-management methods and the capacity gain they bring on NVL72.
Written by AI from the coverage · updated 3 hours ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageGrace Blackwell needs a new scheduler: 27% more usable GPUs from the same racks
SkyPilot 平台通过拓扑感知调度、故障原地恢复和共享集群自动碎片整理，让多节点任务保持机架内本地性，在 Grace Blackwell NVL72 上从同一机架获得 27% 更多可用 GPU。
Heat over time
Heat now 9·Comparable peak 10（Oct 30 01:00）·Comparable change over 24 hours –
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.