A new DLSS 5 GPU mod runs Nvidia’s neural rendering pass on a second graphics card instead of the one doing the actual rendering, recovering the frame rate that pass otherwise costs on a single card. The developer, Marcelo Guibout, published the project as a free ReShade add-on called MGPU Bridge.
The approach mirrors dedicated PhysX GPUs from Nvidia’s older dual-card era: one card renders the game, the other does nothing but the extra processing pass and drives its own monitor. The add-on is free to download from the Neural-coprocessor project on GitHub, though it needs a second display and doubles display latency, since the finished frame has to cross to the second card before it can appear on screen.
How the offload works
Guibout explains that neural rendering happens at the end of the frame: it takes a finished frame and hands a finished frame back. That is what makes it possible to pick up and run somewhere else. MGPU Bridge creates its own D3D12 device on the second GPU, sends each finished frame across, runs DLSS 5’s neural rendering pass there, and displays the result on that card’s own monitor, so nothing has to come back to the rendering card. The render GPU does no neural work at all, and runs cooler for the same reason.
The numbers, from a demo, not a benchmark
The demonstration ran on a Ryzen 7 7800X3D with 32GB of DDR5, two RTX 5060 Ti 16GB cards on PCIe 5.0 x8 each, and a monitor attached to every card. Guibout was clear that the clips, one using a cinematic from The Blood of Dawnwalker and another running Cyberpunk 2077, are technical showcases rather than benchmarks, though he shared frame rate figures anyway. The Blood of Dawnwalker clip is a fitting choice: Nexus Mods opened up open-source Unreal Engine modding tools for the game only recently, and this add-on runs in much the same spirit, built by one developer and shared for anyone to inspect or extend.
In Ultra Performance mode at 1080p, Tom’s Hardware reported that running DLSS 5 on the render card alone produced 69-71fps, while offloading it to the second card lifted that to 157fps, an increase of up to 127%. In DLAA, the gain ran from 44fps to 67-70fps; in Quality mode, from 54-55fps to 91fps; in Performance mode, from 59fps to 106-107fps.
Why this is worth trusting more than a vendor slide
Unlike Nvidia’s own DLSS 5 marketing figures, these numbers come with working code attached. The Neural-coprocessor project lets anyone with a spare GPU and a second monitor run the same test and check Guibout’s figures against their own hardware, rather than take a slide at face value. That is also why the caveats matter as much as the gains: a second display and doubled display latency rule this out for anyone chasing responsiveness over raw frame rate, even if the frame counter looks better.
What to watch next
MGPU Bridge is Guibout’s own project, not an Nvidia feature, so support depends on one developer keeping up with driver and engine changes. Worth watching is whether the technique extends beyond a two-GPU, two-monitor setup, and whether Nvidia folds any of this lesson into a first-party multi-GPU mode of its own.
