Discrete GPU Memory DVFS for Power-Balanced Performance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In discrete graphics products, the memory subsystem consumes a significant portion of the overall device power, reducing the performance of compute and render components due to excessive power consumption.
Innovation Solution
Implementing dynamic voltage and frequency scaling (DVFS) in the memory subsystem of discrete graphics systems based on workload demands, allowing the memory subsystem to adjust its voltage and frequency to optimize performance within the device's power budget.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If the memory subsystem operates at high voltage and frequency to maximize performance, then memory performance is improved, but power consumption increases excessively
Solution Approach 1:
The patent implements dynamic voltage and frequency scaling (DVFS) in the memory subsystem, allowing it to transition between multiple operational states (e.g., STATE 1 with lower voltage/frequency and STATE 2 with higher voltage/frequency) based on real-time workload demands. This dynamic adjustment resolves the contradiction by providing high performance only when needed while consuming less power during lower-demand periods
Solution Approach 2:
The system changes the voltage and frequency parameters of the memory subsystem based on workload characteristics. By monitoring memory access patterns and adjusting voltage/frequency accordingly, the system optimizes the trade-off between performance and power consumption, transitioning parameters only when workload thresholds are crossed
2Productivity
If the memory subsystem consumes more power to improve memory performance, then memory-intensive workloads benefit, but the compute and render engines suffer from reduced available power
Solution Approach 1:
The DVFS mechanism allows the memory subsystem to dynamically adjust its power consumption based on actual memory access patterns. During compute-intensive periods, the memory subsystem can operate at lower power states, freeing up power headroom for the compute and render engines. During memory-intensive workloads, the system transitions to higher performance states only when necessary
Solution Approach 2:
The system incorporates feedback mechanisms that monitor both memory access patterns and overall device power budget. This feedback loop enables intelligent decision-making about when to scale memory performance up or down, ensuring that power allocation balances the needs of different subsystems based on real-time conditions
3Adaptability or versatility
If dynamic voltage and frequency scaling is implemented in the memory subsystem, then power management flexibility is improved, but device complexity increases
Solution Approach 1:
The DVFS implementation is segmented into discrete operational states (e.g., STATE 1, STATE 2) with predefined voltage and frequency combinations. Each state has associated control logic that manages transitions between states, breaking down the complex continuous control problem into manageable discrete steps with clear transition criteria
Data Source
AI summary
In one embodiment, a system on a chip integrated circuit (SoC) is provided that includes graphics processing resources including one or more graphics processing cores a memory subsystem including a memory controller, a physical interface, and a memory device and circuitry to dynamically adjust a voltage and frequency of the memory subsystem based on a workload executed by the graphics processing resources.