How Driver-Level Upscalers Impact Hardware Benchmarking

This article explores how enabling driver-level spatial upscalers, such as AMD Radeon Super Resolution (RSR) and Nvidia Image Scaling (NIS), fundamentally compromises computer hardware benchmarks. While these technologies are great for boosting everyday gaming framerates, leaving them active during hardware testing invalidates the resulting performance data. The following sections break down how driver-level upscaling skews GPU metrics, shifts system bottlenecks to the CPU, introduces algorithmic overhead, and ruins visual parity between competing components.

The Illusion of Native Resolution

Driver-level upscalers work by forcing a game to render at a lower internal resolution (such as 1080p) and stretching that image to fit a higher-resolution display (such as 1440p or 4K), applying a sharpening filter at the final stage.

If a benchmark is intended to measure 4K performance, having RSR or NIS enabled means the hardware is actually only performing the workload of a much lower resolution. This artificially inflates frame rates and makes the graphics card appear far more powerful at heavy workloads than it truly is.

Shifting the Hardware Bottleneck

Accurate benchmarking requires isolating specific components. To test a GPU, you must push the resolution and graphics settings high enough so that the graphics card becomes the limiting factor in the system.

Enabling a spatial upscaler drastically reduces the rendering workload on the GPU. As a result, the framerate increases, and the Central Processing Unit (CPU) must work much harder to process game logic and feed frames to the graphics card. This inadvertently turns a GPU benchmark into a CPU and system memory benchmark, corrupting the data if your goal is to measure graphics performance.

Algorithmic Performance Overhead

Spatial upscaling is not entirely free of performance costs. Both RSR and NIS require a small amount of GPU compute power to apply their scaling and sharpening algorithms to every single frame.

If you compare a game running at native 1080p against the same game upscaled from 1080p to 1440p using a driver-level upscaler, the upscaled version will actually perform slightly worse. This driver overhead usually costs between 1% and 5% of total performance, introducing an unwanted variable when trying to measure the raw, unassisted capability of the silicon.

Loss of Apples-to-Apples Comparisons

The golden rule of hardware benchmarking is maintaining identical workloads across all tested components. Enabling driver-level upscalers breaks this rule in two ways:

Different Algorithms: AMD and Nvidia use different proprietary math to scale and sharpen images, meaning the GPUs are no longer doing the exact same work.
Mismatched Resolutions: The preset scaling percentages in AMD and Nvidia software do not always align. One driver might scale from 77% of the native resolution, while the other scales from 75%, resulting in a different total pixel count being rendered by each card.

If the internal rendering resolutions and scaling methods do not match perfectly, the resulting benchmark numbers cannot be scientifically compared.

Driver-level upscalers operate outside of the game engine. Because of this, built-in game benchmark tools and third-party monitoring software often fail to detect that an upscaler is active.

The software will typically log the benchmark run as being completed at the monitor's native resolution. This creates a massive blind spot in data logging, allowing falsified performance metrics to easily slip into test results without the tester realizing the workload was altered by the GPU driver.

The Verdict for Hardware Testing

For accurate, repeatable hardware benchmarks, driver-level spatial upscalers like RSR and NIS must be strictly disabled. Hardware should always be tested at true native resolutions to ensure the workload is consistent and verifiable. Driver-level upscalers should only be enabled during testing if the specific goal of the benchmark is to evaluate the efficiency and image quality of the upscaling software itself.