{"slug": "skypilot-says-rack-aware-scheduling-makes-27-more-nvl72-capacity-usable", "title": "SkyPilot says rack-aware scheduling makes 27% more NVL72 capacity usable", "summary": "SkyPilot published a technical post on October 29th claiming its rack-aware scheduler makes 27% more capacity usable in NVIDIA NVL72 clusters by reserving two nodes per rack as spares. In SkyPilot's cited model, 16-node job slices yield 2,037 usable nodes across 128 racks, versus 1,602 when jobs consume all 18 nodes per rack. A separate GR00T training test reported 14.4% higher throughput.", "body_md": "# SkyPilot says rack-aware scheduling makes 27% more NVL72 capacity usable\n\n**Its model favors 16-node slices with two spare nodes per rack; a separate GR00T training test reported 14.4% higher throughput.**\n\n        By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff)\n        · Published \n\nPrimary source: [Hacker News](https://skypilot.ai/blog/grace-blackwell-needs-a-new-scheduler)\n\n## Why it matters\n\nNVL72 turns a rack into a high-bandwidth computing domain, so scheduler decisions can affect both job speed and how much capacity survives failures.\n\nSkyPilot says its scheduler can make 27% more capacity usable in NVIDIA NVL72 clusters by leaving two nodes in each rack available as spares. The claim, published in a [technical post on October 29th](https://skypilot.ai/blog/grace-blackwell-needs-a-new-scheduler?ref=runtimewire), is about how operators allocate existing machines, not a hardware upgrade: in the model SkyPilot cites, 16-node job slices yield 2,037 usable nodes across 128 racks, compared with 1,602 when jobs consume all 18 nodes in each rack.…\n\nWhen you purchase through links in our articles, [we may earn a small commission](https://runtimewire.com/affiliate-monetization-standards). This doesn’t affect our editorial independence.", "url": "https://wpnews.pro/news/skypilot-says-rack-aware-scheduling-makes-27-more-nvl72-capacity-usable", "canonical_source": "https://runtimewire.com/article/skypilot-nvl72-topology-aware-scheduling-capacity", "published_at": "2026-10-29 18:06:49+00:00", "updated_at": "2026-10-29 21:17:01.807327+00:00", "lang": "en", "topics": ["ai-infrastructure", "mlops", "ai-chips"], "entities": ["SkyPilot", "NVIDIA", "NVL72", "GR00T"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/skypilot-says-rack-aware-scheduling-makes-27-more-nvl72-capacity-usable", "markdown": "https://wpnews.pro/news/skypilot-says-rack-aware-scheduling-makes-27-more-nvl72-capacity-usable.md", "text": "https://wpnews.pro/news/skypilot-says-rack-aware-scheduling-makes-27-more-nvl72-capacity-usable.txt", "jsonld": "https://wpnews.pro/news/skypilot-says-rack-aware-scheduling-makes-27-more-nvl72-capacity-usable.jsonld"}}