T/TAF 265-2025 智算中心算力池化技术要求和功能测试方法

ICS33.040

CCSM10

团体标准

T/TAF265—2025

智算中心算力池化技术要求和

功能测试方法

Technicalspecificationandfunctionaltestingmethodforcomputing

powerpoolinginartificialintelligencedatacenter

2025-02-10发布2025-02-10实

施

电信终端产业协会发布

T/TAF265—2025

目次

前言................................................................................................................................................................III

引言..................................................................................................................................................................IV

1范围...............................................................................................................................................................1

2规范性引用文件...........................................................................................................................................1

3术语和定义...................................................................................................................................................1

4缩略语...........................................................................................................................................................1

5总体架构.......................................................................................................................................................2

6总体要求.......................................................................................................................................................3

7基础功能要求...............................................................................................................................................3

7.1池化管理...............................................................................................................................................3

7.2池化调度...............................................................................................................................................3

7.3提供任意规格算力资源......................................................................................................................3

7.4虚拟算力资源隔离...............................................................................................................................3

7.5业务热迁移...........................................................................................................................................4

7.6横向扩展...............................................................................................................................................4

8通信网络要求...............................................................................................................................................4

8.1管理网络...............................................................................................................................................4

8.2数据网络...............................................................................................................................................4

9资源池部署与集成要求...............................................................................................................................5

9.1资源池部署...........................................................................................................................................5

9.2资源池集成...........................................................................................................................................5

10兼容性要求.................................................................................................................................................5

10.1网络兼容性.........................................................................................................................................5

10.2GPU设备兼容性...................................................................................................................................5

10.3人工智能芯片API版本兼容性........................................................................................................6

10.4操作系统兼容性.................................................................................................................................6

10.5基础架构平台兼容性.........................................................................................................................6

10.6深度学习框架兼容性.........................................................................................................................6

11可靠性要求.................................................................................................................................................6

12功能测试方法.............................................................................................................................................6

12.1资源固定配置.....................................................................................................................................6

12.2资源动态申请.....................................................................................................................................7

I

T/TAF265—2025

12.3远程调用.............................................................................................................................................7

12.4跨机多卡聚合.....................................................................................................................................7

12.5资源超分.............................................................................................................................................8

12.6资源配额.............................................................................................................................................8

12.7指定卡型号.........................................................................................................................................8

12.8指定节点.............................................................................................................................................9

12.9任务亲和.............................................................................................................................................9

12.10任务队列...........................................................................................................................................9

12.11任务资源空闲超时自动释放.........................................................................................................10

12.12任务运行超时资源自动释放.........................................................................................................10

12.13逻辑资源组.....................................................................................................................................11

12.14自定义资源规格.............................................................................................................................11

12.15双类资源池.....................................................................................................................................11

12.16设备管理.........................................................................................................................................12

12.17节点管理.........................................................................................................................................12

12.18日志收集.........................................................................................................................................12

12.19监控告警.........................................................................................................................................13

12.20组件管理........................................................................................................................................13

12.21统计报表.........................................................................................................................................14

12.22灰度升级.........................................................................................................................................14

12.23任务热迁移.....................................................................................................................................14

12.24显存故障域隔离.............................................................................................................................15

12.25动态扩缩容.....................................................................................................................................15

12.26异构AI算力支持...........................................................................................................................16

12.27调度策略.........................................................................................................................................16

附录A（资料性）能力分级.........................................................................................................................17

参考文献..........................................................................................................................................................19

II

T/TAF265—

2025

前言

本文件按照GB/T1.1—2020《标准化工作导则第1部分：标准化文件的结构和起草规则》的规

定起草。

请注意本文件的某些内容可能涉及专利。本文件的发布机构不承担识别专利的责任。

本文件由电信终端产业协会提出并归口。

本文件起草单位：中国信息通信研究院、北京趋动智能科技有限公司、贵阳市大数据产业有限公司、

中国联合网络通信有限公司、百度在线网络技术（北京）有限公司。

本文件主要起草人：解谦、王鲲、曾晨曦、陈飞、吴金明、马蓁蓁、钟楚然、范灵俊、彭亚松、王

菲、张舒帆、曾楚轩、邓诗贤、陈宁、焦希泉、崔凯。

III

T/TAF265—2025