GPU Deep Sleep Mode via Local Memory Self-Refresh
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing techniques for entering and exiting deep sleep mode in graphics processing units (GPUs) are time-consuming and lead to system performance degradation due to the significant amount of state information that needs to be transmitted between system memory and the GPU, causing inefficiencies in power management.
Innovation Solution
A method where the GPU saves its internal processing state to a locally attached memory system and enters a power-down state, utilizing low power self-refresh modes to efficiently enter and exit deep sleep mode by preserving context within local memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If the GPU enters deep sleep mode by shutting off power to eliminate static power dissipation, then power consumption is reduced, but the time required to enter and exit sleep mode increases due to transmitting state information between system memory and GPU
Solution Approach 1:
The memory system is segmented into local memory devices attached to the GPU and separate system memory. By saving GPU operating state to local memory instead of system memory, the patent eliminates the need to transmit state information across the system bus during sleep mode transitions, thus reducing time loss while maintaining power savings.
Solution Approach 2:
The GPU saves its operating state to local memory devices before entering deep sleep mode. This preliminary action ensures that when the GPU exits sleep mode, the state information is already available in local memory, eliminating the need for time-consuming data transmission from system memory and enabling faster resume operation.
2Reliability
If the GPU transmits significant amounts of state information between system memory and GPU during deep sleep transitions, then operating state is preserved, but system performance degrades due to the time-consuming nature of these transmissions
Solution Approach 1:
The patent extracts the GPU operating state storage function from the system memory and relocates it to local memory devices attached to the GPU. This extraction eliminates the performance-degrading data transmission overhead while maintaining reliable state preservation, as the state information remains locally available without requiring system bus transactions.
Solution Approach 2:
Local memory devices serve as an intermediary between the GPU and system memory. By using this intermediate storage layer, the system can preserve GPU operating state without requiring direct transmission between GPU and system memory during sleep mode transitions, thus maintaining reliability while avoiding performance degradation.
Data Source
AI summary
A technique is disclosed for a graphics processing unit (GPU) to enter and exit a power saving deep sleep mode. The technique involves preserving processing state within local memory by configuring the local memory to operate in a self-refresh mode while the GPU is powered off for deep sleep. An interface circuit coupled to the local memory is configured to prevent spurious GPU signals from disrupting proper self-refresh of the local memory. Spurious GPU signals may result from GPU power down and GPU power up events associated with the GPU entering and exiting the deep sleep mode.