Ai2 优化 GPU 调度以支持高影响力的 AI 研究
ai2 更新 gpu 调度策略，以优先考虑高影响力的研究，改善 150 名内部研究人员的资源分配。ai2 管理数千颗 nvidia h100, b200, 和 b300 gpu，用于大规模分布式 ai 模型训练。
ai2's ai infrastructure team, responsible for providing gpu compute capacity, targeted large, distributed training workloads. they managed thousands of nvidia h100, b200, and b300 gpus arranged in clusters ranging from 88 to 1024 gpus. the team served 150 internal researchers working on ai projects, including llm and vlm training, robotics reinforcement learning, and post-training for scientific agentic use cases. ai2 had demand for gpu time that far exceeded supply, with 2-3x more gpu requests than available. the new scheduling system aims to prioritize high-impact research while maintaining full occupancy.