A new community-developed modification has introduced a method to offload NVIDIA’s demanding DLSS 5 Neural Rendering (DLSS-NR) tasks to a secondary graphics card, effectively reviving the concept of multi-GPU systems to handle modern AI workloads. Released on September 7, 2026, by developer Marcelo Guibout (known online as maohgad-web), the “MGPU Bridge” mod functions as a ReShade add-on that repurposes a second GPU as a dedicated neural coprocessor.

The development follows the official launch of DLSS 5 in NBA 2K27 earlier this September. Unlike previous iterations of Deep Learning Super Sampling that focused primarily on upscaling or frame generation to increase performance, DLSS 5 introduces “Neural Rendering.” This pass enhances lighting, textures, and materials using heavy AI processing, but it carries a significant performance tax. In most scenarios, enabling DLSS 5 results in lower frame rates than running the game with the feature disabled, as the quality-focused neural pass often consumes more than 50% of a single GPU’s available resources.

By offloading this specific “Neural Rendering” stage to a second card, the primary GPU is freed to focus exclusively on rasterization and geometry. In benchmarks conducted with the title The Blood of Dawnwalker, the mod reportedly improved frame rates by between 59% and 127% compared to running DLSS 5 on a single card.

According to reporting by Tom’s Hardware, this approach differs technically from traditional SLI (Scalable Link Interface). While SLI required a physical bridge and synchronized geometry rendering between two cards, MGPU Bridge uses a sequential handoff. The primary card renders the frame, which is then passed to the secondary card’s Tensor cores for the AI enhancement pass.

Despite the performance gains, the current implementation of the mod faces practical hardware limitations:

• Dual-Monitor Requirement: To avoid the latency penalty of sending the processed frame back through the PCIe bus to the primary card, the mod currently requires two monitors. The second monitor must be plugged directly into the secondary GPU to display the final neural-processed output.
• Increased Display Latency: While the frame rate (FPS) increases, the actual display latency is reportedly higher than a single-GPU setup due to the time required for the cross-adapter transfer.
• Sequential Processing: Because the second GPU cannot begin its work until the first has finished rendering the initial frame, there is a hard floor on how much latency can be reduced.

The mod has also been tested on older hardware, including RTX 30 and 40-series cards. However, the performance benefits are less pronounced on older architectures because the DLSS 5 Neural Rendering pass is specifically optimized for the more efficient Tensor cores found in newer Blackwell-based GPUs. While the MGPU Bridge provides a path for enthusiasts with spare hardware to experience high-end AI features, the requirement for a dual-monitor setup and the inherent latency trade-offs remain significant hurdles for general consumer adoption.