SkyPilot has detailed a scheduling approach designed to maximize the benefit of NVIDIA's Grace Blackwell NVL72 racks, which link 72 GPUs via a unified NVLink fabric instead of relying on InfiniBand across nodes. The company says its platform preserves rack-locality for multi-node training jobs through topology-aware placement, in-place recovery from hardware failures, and automatic defragmentation of shared clusters.
skypilot.ai
· 2026-10-29
Nvidia and Oracle announced general availability of Nvidia's Vera Rubin NVL72 racks and standalone Vera CPU instances on Oracle Cloud Infrastructure Superclusters at Oracle AI World. The move follows Google Cloud's similar Vera Rubin availability announcement last week. Separately, independent benchmarks on a Vera server (not run on OCI hardware) showed significantly higher memory bandwidth than a dual Xeon 6 system.
tomshardware.com
· 2026-10-28
Intel announced at OCP 2026 that its previously code-named Crescent Island chip will ship as the GPU D-Series, disclosing a theoretical peak of 1.3 PFLOPS FP8 performance and 1.5 TB/s memory bandwidth. The air-cooled PCIe card runs at 350W TDP with up to 480GB of LPDDR5X memory, built on Intel's new Xe3P architecture, and can scale to eight GPUs per server.
tomshardware.com
· 2026-10-12
GoKawiil