An article published by Igor’s laboratory confirms that Generation of frames from NVIDIA The technology is applied at the end of the frame processing pipeline and fully compatible with FSR 2.0 and Intel XeSS AI scaling pipelines. While there’s no real benefit to doing this it didn’t stop Igor from taking some pioneering cues on what the numbers would be and amazingly it looks like FSR+Frame Generation can actually beat DLSS+Frame Generation in pure performance Figures.
AMD FSR + NVIDIA Frame Generation beats DLSS in NVIDIA RTX 4090 benchmarks
Let’s start with a little background. Frame generation is a technology that is part of NVIDIA’s DLSS 3.0 stack that generates artificially created frames between two frames generated by the AI implementation of DLSS. Think of these frames as padding between two compute-generated frames. It turns out that if you own an NVIDIA 4000-series GPU, you can actually use NVIDIA frame generation with compute-generated frames by Intel XeSS or AMD FSR 2.0.
Image quality between AMD FSR 2.0 + NVIDIA Frame Generation and DLSS + Frame Generation seems quite comparable although NVIDIA Frame Generation and Intel XeSS seem to be on the softer side of things. So without further ado, here are the benchmarks of the three configurations:
Keep in mind that the fps numbers only tell part of the story, to get the full idea we encourage you to go to Igor’s lab and read the frame times, variations and percentile numbers from all three systems, but for now let’s focus on the maximum achievable fps. With DLSS and FG (Frame Generation) you watch a maximum value of 224 fps. Pretty decent right? Keep in mind that even this value can be seriously throttled by software due to frame time limitations in the game engine.
Then we have AMD FSR which eliminates frames in the back-end and NVIDIA Frame Generation is used to fill in the gaps. The maximum achievable FPS here in Spider-Man Remastered is 231.4 fps, which is slightly higher than the native approach of using DLSS. That said, this appears to be a software bottleneck situation as other metrics are actually quite comparable to the DLSS-based implementation.
Finally, we have the Intel XeSS + NVIDIA Frame Generation based runtime which results in a maximum achievable fps of 203.9 – which doesn’t appear to be a software bottleneck. As noted above, Intel XeSS is also slightly softer in the image quality department than FSR 2.0 or NVIDIA DLSS.
One thing is for sure though, further testing in titles that support this will be needed to verify if FSR could actually boost performance when paired with NVIDIA Frame Generation on RTX 4000 series GPUs – in a title that encounters no software bottlenecks. Vendor compatibility between scaling technologies is something quite exciting and could mean the best of both worlds for consumers.