近一半的企业不顾成本问题,在未充分利用的GPU上浪费了数以百万计的资金,促使ClearML等工具通过分数性GPU共享来提高效率。 Nearly half of enterprises waste millions on underused GPUs despite cost concerns, prompting tools like ClearML to boost efficiency via fractional GPU sharing.

一份新的ClearML报告显示,尽管2025-2026年将成本控制和效率列为优先事项,但由于GPU能力利用不足,近一半的企业正在浪费数百万美元。 A new ClearML report reveals that nearly half of enterprises are wasting millions due to underutilized GPU capacity despite prioritizing cost control and efficiency in 2025–2026. 35%的企业希望提高GPU的利用率,44%的企业依赖于手动分配工作量或缺乏正式的策略,导致人工智能开发的延迟. While 35% aim to improve GPU utilization, 44% still rely on manual workload assignment or lack formal strategies, creating delays in AI development. 成本管理是53%的最高挑战,而数据、模型和计算治理是许多人的一个关键优先事项。 Cost management is the top challenge for 53%, and governance of data, models, and compute is a key priority for many. 为了解决低效率问题,ClearML扩大了对AMD Intinct GPU的分数式GPU分割的支持,使一个带有自动化中央管理的单一GPU能够同时处理多重工作量。 To address inefficiencies, ClearML has expanded support for fractional GPU partitioning on AMD Instinct GPUs, enabling multiple workloads to run simultaneously on a single GPU with automated, centralized management. 硅不可知平台提高了资源效率,减少了闲置能力,并支持了多种多样的环境——帮助企业在不增加基础设施成本的情况下最大限度地实现ROI。 The silicon-agnostic platform improves resource efficiency, reduces idle capacity, and supports heterogeneous environments—helping enterprises maximize ROI without increasing infrastructure costs.