Apex Compute has introduced Apex Neural Rendering (ANR), an AI-powered graphics technology that the company claims delivers faster frame generation, better image quality and lower GPU memory consumption than NVIDIA DLSS Frame Generation. Its published benchmarks show ANR generating 4K frames in 1.01 milliseconds compared with 1.26 milliseconds for DLSS, while using substantially less video memory.
The technology combines three components: ANR Frame Generation (ANR FG), ANR Super Resolution (ANR SR) and ANR Ray Reconstruction (ANR RR). Apex Compute has released a public beta of its frame-generation system, while super-resolution upscaling and ray reconstruction are planned for later beta releases.
Unlike NVIDIA’s proprietary DLSS Frame Generation implementation, ANR is designed to operate across multiple GPU manufacturers, including NVIDIA GeForce, AMD Radeon and Intel Arc graphics cards.
Apex Compute published its performance and image-quality comparisons on its official Neural Rendering website, alongside demonstration images and footage from The Last of Us Part II. However, the benchmarks were produced by the developer itself and have not been independently validated.
In the company’s testing, conducted using an NVIDIA GeForce RTX 5090 at 3840 × 2160 resolution, ANR FG completed each generated frame in 1.01 milliseconds at 2X frame generation. NVIDIA DLSS Frame Generation required 1.26 milliseconds under the reported comparison conditions.
At 4X frame generation, Apex Compute recorded an average processing time of 0.61 milliseconds per generated frame, compared with 0.82 milliseconds for DLSS. The 4X measurement represents the average time across three generated frames rather than the total processing time for the complete sequence.
GPU memory consumption represents another notable difference. According to Apex Compute, ANR FG uses 214 MB of GPU memory at 4K, while the compared DLSS Frame Generation implementation requires 562 MB. That represents approximately 62% lower memory consumption in the company’s measurements.
Apex Compute also states that its frame-generation neural network weights occupy just 0.17 MB, suggesting a comparatively compact model.
These figures measure frame-generation processing overhead rather than complete game performance. They do not establish that ANR delivers higher overall frame rates, lower end-to-end input latency or better performance across different GPUs and games. The company has also not established that its results apply to every DLSS version or configuration.
Beyond performance, Apex Compute claims its system produces more accurate generated frames than NVIDIA DLSS.
The company evaluated both technologies using eight rendered 4K clips and three image-quality metrics: CGVQM-5, ColorVideoVDP and peak signal-to-noise ratio (PSNR).
At 2X frame generation, ANR FG received a CGVQM-5 score of 96.59 compared with 93.66 for DLSS. Its ColorVideoVDP result was 9.24 versus 8.77, while PSNR reached 33.84 dB compared with 31.22 dB for NVIDIA’s implementation.
The reported advantage continued across higher frame-generation multipliers. At 4X, ANR scored 94.20 on CGVQM-5 compared with 89.21 for DLSS, while its PSNR result reached 34.50 dB against 31.84 dB.
Apex Compute’s results also show ANR leading across the same three metrics at 3X, 5X and 6X frame generation.
The company has published side-by-side comparisons featuring fast-moving particles, spinning objects, smoke and scenes from The Last of Us Part II. These demonstrations are intended to illustrate how the two systems handle motion, overlapping objects and newly revealed image regions.
However, higher scores in the selected tests do not automatically establish superior image quality in every game. Independent evaluations using a wider range of scenes, graphics settings and DLSS configurations will be necessary to determine how consistently ANR performs.
One of ANR’s most significant features is its broad hardware compatibility.
According to Apex Compute’s official documentation, the frame-generation beta supports NVIDIA GeForce GTX 16-series and RTX 20-, 30-, 40- and 50-series graphics cards. AMD compatibility extends from Radeon RX 5000 through RX 9000 GPUs, including Steam Deck hardware, while Intel Arc A-series and B-series GPUs are also supported.
The beta operates on Windows 10 and Windows 11, as well as Linux and SteamOS through Proton. Apex Compute additionally lists Apple silicon M1 and newer hardware, with macOS 26 or later support.
This broader compatibility could make neural frame generation accessible to PC gamers using graphics cards that cannot run NVIDIA’s official DLSS Frame Generation implementation.
The current beta is distributed through Apex Compute’s GitHub repository and uses OptiScaler to integrate with supported games.
Users can download either a complete package containing the required OptiScaler components or an add-on intended for existing OptiScaler installations. Installation involves placing the necessary files into a game’s directory and enabling its frame-generation option.
Because the implementation relies on existing game integration, it is not a universal frame-generation solution for every PC title. Compatibility and performance may vary depending on the game, GPU, driver and operating system.
Apex Compute has also provided a recording and comparison utility for capturing game frame-generation inputs, allowing different implementations to be evaluated using the same recorded data.
For players following the development of neural graphics technologies, ANR arrives as established solutions continue evolving. NVIDIA’s DLSS 5 rollout is expanding support for its latest rendering technologies, while games such as Anno 117: Pax Romana have introduced DLSS, AMD FSR and Intel XeSS rendering options.
ANR’s approach is particularly interesting because it targets several GPU architectures rather than restricting frame generation to one manufacturer’s latest hardware.
Nevertheless, the current release remains an experimental beta focused on frame generation. ANR Super Resolution and ANR Ray Reconstruction are not yet available in the same form, and Apex Compute has not announced firm public release dates for those components.
The company’s performance and image-quality claims are supported by its published measurements and demonstration material, but independent testing is still required before ANR can be considered demonstrably faster or visually superior to NVIDIA DLSS in real-world gaming.
