GPU in Furmark, Superpostion, OCT, Valley all pull max power but in games it’s less than half?

Understanding Power Discrepancies in Modern GPUs: A Case Study with MSI RX 6800 Gaming X Trio

Introduction

Graphics Processing Units (GPUs) are complex components, and their performance characteristics can sometimes be confusing, especially when monitoring tools reveal inconsistent data across different scenarios. Recently, some users have observed that their GPUs draw maximum power during synthetic benchmarks but less than half that during actual gaming sessions. This blog post explores a recent case involving an MSI RX 6800 Gaming X Trio 16GB to understand the underlying reasons behind such discrepancies.

Scenario Overview

The user acquired a used MSI RX 6800 Gaming X Trio GPU primarily intended for cryptocurrency mining. After thorough cleaning and servicing—including replacing thermal pads and paste—the card was functional in Windows with stable temperatures during synthetic testing. Key observations included:

Synthetic Benchmarks: Furmark, Superposition 8K, and Valley benchmark consistently pushed the GPU to draw maximum power, often exceeding 200W.
Gaming Performance: When running games like Cyberpunk 2077, Spiderman 2, and others, GPU power consumption measured at roughly half that of synthetic tests (~100W), despite the GPU clock speeds remaining high (near maximum according to monitoring tools).

The Question

Why does the GPU hit maximum power in synthetic benchmarks but draw significantly less during real-world gaming, leading to reduced performance and utilization?

Potential Causes and Explanations

Power Limit and BIOS Settings
Custom BIOS flashing or factory settings might impose different power profiles for synthetic versus gaming workloads.
Overclocking or adjusting power limits in AMD Adrenaline may not always be fully enforced during gaming, especially if thermal or power constraints are detected dynamically.
Power Management Behavior
Modern GPUs dynamically adjust power and clock speeds based on workload, thermal conditions, and power headroom.
Synthetic benchmarks often simulate maximum load steadily, causing the GPU to stay at or near its maximum power draw.
Games introduce varying workloads with potential dips in load, especially during scenes with less intensive rendering, resulting in lower overall power consumption.
Monitoring Tools and Reporting Discrepancies
Software like HWInfo, MSI Afterburner, and AMD Overlay may report maximum clock speeds or GPU states that are not constantly active.
In-game power consumption can appear lower if the GPU is modulation its power based on current demand, even if clocks are high.
Power Throttling and Dynamic Boost Technologies
GPUs feature dynamic features that optimize power usage;