GFX Core Power Gating via Subslice Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing graphics processing unit (GFX) cores consume unnecessary power when operating in the RC0 active state due to the need to transition the entire core from a low-power RC6 state back to active mode for light workloads, leading to increased energy consumption and reduced battery life.
Innovation Solution
Implementing a method to power gate GFX cores by transitioning from a 16 EU/2 Sampler mode to an 8 EU/1 Sampler mode, allowing for RC6 state residency and reducing energy consumption while maintaining performance by selectively powering down Subslices and Samplers based on workload analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If the entire GFX core transitions to RC6 power-down state, then energy consumption is reduced, but the entire core must transition back to RC0 active state if any portion becomes active, causing unnecessary power consumption
Solution Approach 1:
The GFX core is divided into multiple independent subslices, each capable of entering RC6 state independently. This segmentation allows partial power-down where only active subslices remain in RC0 state while inactive ones transition to RC6, avoiding the all-or-nothing transition requirement of the prior art.
Solution Approach 2:
The system dynamically adjusts the power state of individual subslices based on real-time workload analysis. The GFX core can transition between different operational modes (full active, partial active with some subslices in RC6, full RC6) depending on the current processing demands, enabling adaptive power management.
2Productivity
If the GFX core operates in RC0 active state, then processing capability is maintained, but energy consumption increases for light workloads
Solution Approach 1:
Instead of keeping the entire GFX core active in RC0 state for light workloads, the system activates only the necessary subset of subslices required to handle the current workload. The remaining subslices are placed in RC6 state, providing partial action that maintains sufficient processing capability while reducing overall energy consumption.
Solution Approach 2:
The system changes the operational parameters of the GFX core by transitioning individual subslices between RC0 and RC6 states based on workload characteristics. This parameter change enables the core to adapt its power consumption level to match the actual processing requirements, rather than operating at a fixed high-power state.
Data Source
AI summary
A method includes executing a workload on a graphics (GFX) core in a first mode the GFX core comprising a plurality of Subslices wherein each of the plurality of Subslices dissipates power. The method further includes calculating a number of clock cycles, Tfirst mode, required for the GFX core to perform the workload in the first mode during a first decision window comprising a plurality of clock cycles and calculating a number of clock cycles, Tsecond mode, required for the GFX core to perform the workload in a second mode during the first decision window wherein the second mode comprises executing the workload with fewer of the plurality of Subslices receiving power than when executing the workload in the first mode. It is then determined, based in part upon Tfirst mode and Tsecond mode, if an energy savings is possible by transitioning the GFX core to the second mode.