Apex Neural Rendering (ANR) Debuts as Cross-Vendor DLSS Frame Generation Alternative
Proprietary upscaling and frame interpolation technologies have long divided PC gamers along hardware lines, locking flagship features behind specific GPU architectures. Aiming to break that exclusivity, Apex Compute has unveiled Apex Neural Rendering (ANR)—a cross-platform neural rendering suite encompassing Frame Generation (ANR FG), Super Resolution (ANR SR), and Ray Reconstruction (ANR RR). Available now in public beta via GitHub, the initial release focuses on frame generation and claims faster per-frame execution, significantly lower VRAM overhead, and superior image quality compared to NVIDIA DLSS Frame Generation.
Key takeaways
- Cross-Vendor GPU Compatibility: Unlike proprietary solutions restricted to specific tensor cores, ANR FG supports NVIDIA GeForce (GTX 16 and RTX 20–50 series), AMD Radeon (RX 5000–9000 series and Steam Deck), Intel Arc (A-series and B-series), and Apple Silicon (M1 and newer via Metal).
- Ultra-Compact Neural Footprint: Apex reports that ANR FG's neural network weights occupy just 0.17 MB, consuming only 214 MB of VRAM at 4K compared to 562 MB for NVIDIA DLSS Frame Generation.
- Faster Per-Frame Generation: Internal benchmarks on an RTX 5090 at 3840 × 2160 show ANR FG generating frames in 1.01 ms (2X) and 0.61 ms (4X), outpacing DLSS Frame Generation (1.26 ms and 0.82 ms).
- OptiScaler Integration & Roadmap: The beta ships both as a standalone injector and an OptiScaler add-on for Windows 10/11, Linux, and SteamOS (Proton). Super Resolution, Ray Reconstruction, and full source code are slated to follow soon.
How Apex Neural Rendering Challenges Proprietary Frame Generation
Modern frame generation relies on optical flow and motion vectors to synthesize intermediate frames, boosting perceived smoothness in CPU- or GPU-bound titles. However, existing implementations either require newer-generation hardware (such as RTX 40/50 or RX 9000 GPUs) or introduce noticeable VRAM penalties and ghosting artifacts on older cards.
What distinguishes ANR's architecture and benchmark claims?
Apex Compute states that ANR rebuilds neural rendering around a compact, vendor-agnostic model architecture:
- Efficiency & Latency: On an RTX 5090 at 4K, ANR FG reportedly cuts per-frame generation overhead by roughly 20–25% versus DLSS FG while slashing memory footprint by over 60%. On Apple's M2 Max using native Metal, 1080p 2X frame generation clocks in at 4.26 ms per generated frame.
- Image Quality Metrics: Across test scenes from The Last of Us Part II Remastered and Hogwarts Legacy, Apex's published tables show ANR scoring ahead of DLSS in CGVQM-5, ColorVideoVDP, and PSNR evaluations across 2X through 6X multi-frame modes.
- Competitive AMD & Intel Scaling: First-party charts indicate ANR outperforms AMD FSR 3.1 across tested modes on the Radeon RX 7800 XT, Intel Arc B580, and Arc A770, while trading blows with FSR 4 ML on the RX 9060 XT.
ANR FG vs. NVIDIA DLSS Frame Generation Overview
| Specification / Benchmark Metric | Apex Neural Rendering (ANR FG) | NVIDIA DLSS Frame Generation |
|---|---|---|
| Supported Hardware | GTX 16 / RTX 20–50, RX 5000–9000, Steam Deck, Arc A/B, Apple M1+ | NVIDIA GeForce RTX 40 and RTX 50 series |
| 4K Per-Frame Time (RTX 5090, 2X / 4X) | 1.01 ms (2X) / 0.61 ms (4X) | 1.26 ms (2X) / 0.82 ms (4X) |
| 4K GPU Memory Footprint | 214 MB (0.17 MB network weights) | 562 MB |
| OS & API Support | Windows 10/11, Linux/SteamOS (Proton), macOS 26+ (Metal) | Windows (DirectX 12 / Vulkan) |
Frequently asked questions
Is Apex Neural Rendering completely open source right now?
While the beta binaries, installer scripts, and documentation are hosted publicly on the apex-compute/ANR GitHub repository, Apex Compute notes that full source code release and licensing details are still forthcoming alongside the remaining upscaling models.
How do gamers install and enable ANR Frame Generation?
Users extract the beta package into a game's executable directory and run install.cmd on Windows or install.sh on Linux/Steam Deck (with Proton DLL overrides). Because it hooks via OptiScaler, gamers simply select the standard NVIDIA Frame Generation toggle inside the game's graphics menu.
When will ANR Super Resolution and Ray Reconstruction launch?
Apex Compute states that both ANR SR (upscaling) and ANR RR (ray-tracing denoising) are currently in active model training and will enter public beta in upcoming updates.
Bottom line
If independent third-party testing validates Apex Compute's latency and image quality claims, ANR could become a major equalizer across PC and handheld gaming once its full suite and source code arrive.