Back to articles
DevPickView feed
Hacker NewsHN submitter: wondersky
Original published:

Grace Blackwell needs a new scheduler: 27% more usable GPUs from the same racks

Grace Blackwell needs a new scheduler: 27% more usable GPUs from the same racks
AI summary
Editorial policy

SkyPilot explains how topology-aware placement and automatic defragmentation improve GPU utilization by 27% in NVL72 environments.

Key points

  • •NVL72 connects all 72 GPUs in a rack to a single NVLink fabric, but the speedup is limited to rack-local jobs.
  • •SkyPilot maintains multi-node training and inference on NVLink through topology-aware placement.
  • •In-place recovery from GPU failures and automatic defragmentation allow for 27% more usable GPUs from the same racks.
Read the original(opens in a new tab)
#Semiconductors & Infrastructure#SkyPilot#NVL72

Related articles

See more
Cloud Watch

Toshiba Launches GridDB E.E. V6.0, a Database for IoT Systems with Integrated Data Collection and Analysis Features

Toshiba launched GridDB E.E. V6.0, a time-series database for large-scale IoT systems that enhances real-time processing of petabyte-scale data and development ease via SQL.

ITmedia

Google to invest $17 billion in Wyoming data centers, promising to cover all electricity costs

Google announced a $17 billion investment in Wyoming data centers, promising to cover all electricity costs and introduce eco-friendly power generation to address local opposition.

The Verge

OpenAI is stepping back from NDAs for data center projects

Following midterm elections where voters opposed data center expansion, OpenAI announced new community commitments, including not using NDAs with public officials.

Cloud Watch

Fixstars and Kokusai Chotatsu Joho to provide high-speed inference AI services via domestic GPU servers using 'FastMetal'

Fixstars partnered with International Procurement Information to launch a gateway service that enables fast, cutting‑edge AI on domestic GPU servers.

INTERNET Watch

Akamai Launches AI Inference Service 'Akamai Inference' and AI Agent Execution Environment 'Akamai Agent Sandbox'

Akamai announced new products providing an AI inference environment and an isolated execution environment for agents on its cloud service, supporting enterprises from model building to operation.