Innovative Multi-GPU Setup Splits Rendering and Neural Workloads for Enhanced Gaming Performance

For those who remember the golden era of PC gaming—when SLI configurations and dedicated PhysX cards were the norm—a recent breakthrough by modder Marcelo Guibout brings a fresh twist to multi-GPU setups. In a demonstration shared on YouTube, Guibout showcased a gaming rig equipped with two NVIDIA RTX 5060 Ti 16 GB graphics cards, each handling distinct tasks: one focused on traditional rendering, while the other managed Neural Rendering workloads.

How the Multi-GPU Neural Rendering Setup Works

This approach echoes the days of dedicated PhysX cards, where specific hardware was assigned to offload certain processing tasks. By dividing the rendering and Neural Rendering duties between two GPUs, the primary graphics card can dedicate more resources to rasterization or ray tracing, rather than sharing capacity with demanding DLSS 5 Neural Rendering processes.

Guibout utilized the Neural Coprocessor (also known as MGPU Bridge), a publicly available add-on for ReShade, to enable this multi-GPU processing. The setup was tested in popular titles such as Cyberpunk 2077 and The Blood of Dawnwalker. The process involves the primary GPU rendering the frame, which is then transferred via the PCIe link to the secondary GPU. This second GPU applies the DLSS 5 Neural Rendering pass before outputting the final frame to the display.

Performance Gains and Limitations

According to Guibout’s benchmarks, enabling the Neural Coprocessor in this dual-GPU configuration resulted in frame rate improvements ranging from 59% to 125% compared to running DLSS 5 on a single GPU. However, it’s important to note that overall performance was still lower than running with DLSS 5 disabled entirely. The add-on can be fine-tuned through the same ReShade control panel used by DLSS 5 Swapper, allowing users to balance performance and image quality based on their preferences.

This experiment highlights the potential of leveraging multiple GPUs for specialized tasks in modern gaming PCs. While not yet a mainstream solution, it demonstrates how creative hardware configurations can push the boundaries of real-time graphics and neural rendering technologies.