How Resizable BAR and SAM Negatively Impact Game Benchmarks

Resizable BAR (ReBAR) and AMD's Smart Access Memory (SAM) are PCIe technologies designed to boost gaming performance by allowing the CPU to access the entire GPU frame buffer at once, rather than in traditional 256MB chunks. While this feature generally improves frame rates, hardware benchmarking reveals that it actually degrades performance in several specific gaming titles. This article explains the technical reasons behind these negative performance impacts, examining how game engine incompatibilities, memory management overhead, PCIe bus contention, and driver handling cause lower average frame rates and severe micro-stuttering in certain benchmarks.

Game Engine Memory Allocation Flaws

The primary reason ReBAR and SAM hurt performance in specific titles comes down to how a game engine is coded to handle video memory (VRAM). For decades, developers built game engines around the assumption that the CPU could only view 256MB of VRAM at any given moment.

When ReBAR or SAM removes this limit, older or poorly optimized game engines do not know how to efficiently manage the newly available multi-gigabyte memory pool. Instead of streaming assets smoothly, the engine's memory allocator can become overwhelmed. This inefficiency creates a bottleneck, delaying the delivery of critical rendering data to the GPU and resulting in lower overall frame rates during a benchmark run.

Increased CPU Overhead

Enabling full access to the GPU frame buffer shifts additional workload onto the CPU. In games that are not specifically optimized for ReBAR or SAM, the CPU must spend extra cycles managing the larger memory address space and tracking asset locations in the VRAM.

In hardware benchmarking, this is particularly noticeable when testing graphics cards at lower resolutions like 1080p, where the system is already CPU-bound. The extra CPU overhead introduced by SAM or ReBAR steals processing power away from the game's core logic and draw calls. Consequently, benchmarks in CPU-heavy titles like Counter-Strike 2 or Watch Dogs: Legion have historically shown performance regressions when the feature is active.

PCIe Bus Contention and 1% Low Drops

In benchmarking, average FPS is only part of the data; 1% and 0.1% low frame rates are critical for measuring frame pacing and smoothness. ReBAR and SAM can heavily degrade these metrics by causing PCIe bus contention.

Without ReBAR, data transfers across the PCIe lanes are small, predictable, and tightly queued. With ReBAR enabled, a poorly optimized game might attempt to push massive, unoptimized batches of data across the PCIe bus all at once. This floods the PCIe lanes and delays time-sensitive rendering instructions. During a benchmark, this congestion manifests as sudden frame time spikes, dragging down the 1% and 0.1% lows and causing visible micro-stuttering, even if the average frame rate appears unaffected.

API Limitations in DirectX 11

While DirectX 12 and Vulkan are low-level APIs designed to handle modern memory management, DirectX 11 handles memory allocation largely through the graphics driver rather than the game engine itself.

When benchmarking DirectX 11 titles with SAM or ReBAR enabled, the graphics driver has to translate the older API's memory requests into a format that utilizes the expanded address space. This translation layer often introduces latency. As a result, many older DirectX 11 benchmarking staples see either zero benefit or a slight performance penalty when ReBAR is turned on.

Driver Whitelisting vs. Global Activation

The negative impact of this technology on benchmarking is so prevalent that GPU manufacturers handle it differently at the driver level.

NVIDIA utilizes a strict driver whitelist for Resizable BAR. If a game is known to suffer from memory allocation bugs or stuttering with ReBAR enabled, NVIDIA disables the feature for that specific title at the driver level, even if the user has it turned on in the BIOS. For example, titles like Red Dead Redemption 2 and Cyberpunk 2077 have required specific driver profile adjustments to prevent ReBAR from hurting performance.

Conversely, AMD’s Smart Access Memory is more aggressively applied across the board. While this often yields higher gains in well-optimized titles, it also means that benchmarking an AMD system can reveal sudden, drastic performance regressions in incompatible games, forcing hardware testers to manually disable SAM in the BIOS or software suite to get an accurate baseline of the hardware's capability.