发明名称 METHODS AND APPARATUS TO SUPPORT DYNAMIC ADJUSTMENT OF GRAPHICS PROCESSING UNIT FREQUENCY
摘要 Disclosed methods support dynamic adjustment of graphics processing unit (GPU) frequency. According to one embodiment, a program comprises workload to execute, at least in part, on a GPU of a data processing system. A predetermined memory/compute ratio for the program is automatically retrieved, in response to the program being called for execution. The memory/compute ratio represents a ratio of memory accesses within the program, relative to compute operations within the program. In addition, a frequency of the GPU is automatically adjusted, based on the predetermined memory/compute ratio for the program. For instance, the GPU may be set to a relatively low frequency if the predetermined memory/compute ratio is relatively high, or to a relatively high frequency if the predetermined ratio is relatively low. After the frequency of the GPU is automatically adjusted, the program may execute, at least in part, on the GPU. Other embodiments are described and claimed.
申请公布号 US2017076422(A1) 申请公布日期 2017.03.16
申请号 US201415122108 申请日期 2014.03.27
申请人 Shen Lei;Lueh Guei-Yuan;Jiang Yong;lntel Corporation 发明人 Shen Lei;Lueh Guei-Yuan;Jiang Yong
分类号 G06T1/60;G06F9/50;G06T1/20;G06F1/32 主分类号 G06T1/60
代理机构 代理人
主权项 1. A data processing system with support for dynamic adjustment of graphics processing unit frequency, the data processing system comprising: a central processing unit (CPU); a graphics processing unit (GPU) responsive to the CPU; a machine accessible medium responsive to the CPU; and data in the machine accessible medium which, when accessed by the CPU, enables the data processing system to perform operations comprising: in response to a program being called for execution, automatically retrieving a predetermined memory/compute ratio for the program, wherein the program comprises workload to execute at least in part on the GPU, and wherein the predetermined memory/compute ratio represents a ratio of memory accesses within the program, relative to compute operations within the program; andautomatically adjusting a frequency of the GPU, based on the predetermined memory/compute ratio for the program.
地址 Shanghai CN