摘要
为了解决GPU因高并行性导致高功耗从而需要昂贵的冷却解决方案的问题,提出了一种层次化的GPU功耗模型,该模型的第一层是实测层,第二层是核心层,第三层是测试应用程序和仿真工具层。实验结果显示,GPU层次功耗模型框架在脱机状态下,近似随机分布的数据流量突发性相当好,为今后的GPU设计提供了理论和实践的指导。
To address the GPU because of the high parallelism and thus lead to high power loss requires expensive cooling solutions,this paper proposes a hierarchical GPU power loss model,the first layer of the model is the actual layer,its second layer is the core layer the third layer is the test application and simulation tool level.Experimental results show that,GPU-level power modeling framework in the offline state,similar to random distribution of data traffic sudden very good for the future of GPU design provides theoretical and practical guidance.
出处
《电脑开发与应用》
2011年第8期53-54,57,共3页
Computer Development & Applications
