According to a company called Apex Compute, there is a new player in neural rendering. The company has developed Apex Neural Rendering, which reportedly offers higher quality and performance than NVIDIA's DLSS implementations. It includes three key components: frame generation (ANR FG), super resolution upscaling (ANR SR), and ray reconstruction denoising for ray tracing (ANR RR). Running frame generation on NVIDIA's GeForce RTX 5090 at 4K, the ANR FG achieves a latency of 1.01 ms per generated frame at a 2x setting, while NVIDIA's DLSS Frame Generation records 1.26 ms. We are not sure which version of DLSS FG was tested, but we can assume it was the latest version.

Regarding memory consumption, ANR FG uses only 214 MB of GPU memory, while DLSS FG requires more than double that, at 562 MB. Reportedly, the ANR FG neural network weights fit within just 0.17 MB, which is incredibly efficient. Based on samples on the company's website, the image quality that ANR delivers matches DLSS and even surpasses it in some cases. Apex Compute also managed to get Hogwarts Legacy and The Last of Us Part II Remastered working, showing the difference compared to DLSS, resulting in frame generation with more visual fidelity. ANR FG is a beta injector built around OptiScaler, which works in any game offering NVIDIA DLSS Frame Generation. You can view the results below.Supported GPUs include all of NVIDIA's GeForce RTX 20 series and newer, as well as the GeForce GTX 16 series. On the AMD side, support starts from the Radeon RX 5000 series and extends to the Steam Deck. Intel's Arc A-series and B-series are also supported, and surprisingly, Apple's M1 and later are compatible as well. The operating systems tested and supported include the now end-of-life Windows 10, Windows 11, Linux, SteamOS with Proton, and macOS 26 or later. Benchmarks comparing quality, image noise, and more are showcased below.Finally, Apex Compute will open-source this ANR framework with its components soon, which will be available on GitHub.