Can Frame Generation Techniques Like LSFG Enhance Dual-GPU AI Upscaling Architectures?

As a follow-up to my previous post about a dual-GPU rendering concept—where one GPU (e.g., AMD RX 9070 XT) handles real-time 1440p or 4K rasterization, while a second dedicated AI accelerator (like an NVIDIA Tesla T4 or Ada L40) performs real-time upscaling to 8K—I’ve been thinking about another layer of enhancement: AI-based frame generation.

Specifically, tools like LSFG (Lossless Scaling Frame Generation), a third-party software-based frame generation technique, could potentially fit into this model. LSFG works by injecting interpolated frames using motion estimation to boost frame rates—without re-rendering them on the main GPU. This makes it a compelling addition in a split pipeline where the raster GPU and the AI accelerator are already dividing tasks.

In theory, the AI GPU could take on not just upscaling but also real-time frame generation. With fast data exchange through PCIe Gen 5 or NVLink, and efficient runtimes like TensorRT or ONNX, this hybrid process might work within a low 3–6ms latency window per frame. The result: sharper, smoother, and more fluid visuals—without the raster GPU ever touching 8K or high refresh workloads.

Combining AI super-resolution with frame interpolation could be a significant performance multiplier. It offloads work from the primary GPU and potentially reduces system-level latency. But, of course, the idea still faces real-world challenges like frame sync, memory sharing, and inter-vendor hardware compatibility. Nonetheless, the components—runtimes, hardware, and frameworks—already exist.

I'm curious what professionals working in graphics pipelines, game engines, or real-time video AI think:
Could something like LSFG be effectively integrated into a dual-GPU upscaling framework? Could this help push toward ultra-high-resolution and ultra-high-framerate rendering without brute-force GPU horsepower?

If you’re working in this space, I’d love to hear your insights. Just sharing this as a technical curiosity.
Feel free to drop your thoughts at [e-mail]
— Muhammed Shafin P.