为什么 AI 任务不会简单地转移到电力最便宜的地方

发生了什么

雷锋网探讨了将 AI 工作负载转移到电力供应更充足或价格更低的地区所面临的技术和市场障碍。将任务迁移到其他集群，需要能够使用目的地的硬件，并提前准备好模型、数据和软件；价格也很重要，因为调度器看到的可能是计算资源租赁费用，而非电网的实时电价。宗纪科技 CEO 付智表示，公司自 2023 年起一直在探索分布式计算。

接下来读什么

• 今日还有哪些重要变化Hugging Face 提议为 Transformers 制定模型集成契约

这件事到目前为止

Leiphone examined the technical and commercial hurdles to shifting AI workloads toward regions with more available or cheaper power. Moving a task across clusters requires access to the destination hardware and advance preparation of its models, data, and software. A scheduler may also see compute rental costs rather than the grid’s real-time electricity price, complicating decisions based on power costs. Zongji Technology CEO Fu Zhi said the company has explored distributed computing since 2023.