Astonishing mod runs DLSS 5 on a second GPU to boost neural-rendered FPS up to 127% — game renders on one card, neural post-processing runs on the other, much like dedicated PhysX GPUs
- ID
- 22079
- Status
- summarized
- Published
- 07 Sep 2026, 8:33 PM
- Fetched
- 07 Sep 2026, 8:42 PM
- Provider
- Tom's Hardware
- Category
- technology
- Original URL
- https://www.tomshardware.com/pc-components/gpus/astonishing-mod-runs-nvidia-dlss-5-on-a-second-gpu-using-a-reshade-add-on-to-reduce-performance-impact-boosts-neural-rendered-fps-up-to-127-percent-game-renders-on-one-card-neural-post-processing-runs-on-the-other-much-like-dedicated-physx-gpus
- Source URL
- https://www.tomshardware.com/feeds/all
Summary
- Score
- 3.5
- Created
- 07 Sep 2026, 8:46 PM
- Tags
- Audience
- developersai_ml_learners
What happened
A mod uses a ReShade add-on to offload DLSS 5 neural post-processing onto a second GPU while the primary card handles game rendering, boosting neural-rendered FPS by up to 127%. The approach mirrors the old dedicated PhysX GPU pattern, splitting the rendering pipeline across two cards.
Why it matters
The interesting architectural takeaway is splitting a neural inference workload onto a separate GPU to avoid competing for the primary card's compute and VRAM — a pattern directly relevant to anyone building multi-GPU AI inference pipelines or edge deployments where you might dedicate one accelerator to a model and another to application logic.
Discussion angle
Does the 'dedicated accelerator for a specific neural task' pattern from this mod translate to production AI workloads — e.g., running a vision model on a secondary GPU while the primary serves an LLM — or is this only viable in single-machine gaming setups?