Apex Compute Debuts Neural Rendering Beta

Apex Compute introduced Apex Neural Rendering (ANR), a cross-platform neural rendering suite that the company says can generate frames up to 2.4X faster than NVIDIA DLSS while using 62% less GPU memory. Its Frame Generation component is available in beta; Super Resolution and Ray Reconstruction remain in development. The beta supports Windows 10/11, Linux, and SteamOS through Proton. A separate native Metal test kit supports Apple silicon on macOS 26 and newer.

ANR supports NVIDIA GeForce GTX 16 and RTX 20/30/40/50 Series, AMD Radeon RX 5000/6000/7000/9000 Series and Steam Deck, plus Intel Arc A-Series and B-Series. In Apex Compute’s internal RTX 5090 tests at 3840×2160, ANR took 1.01 milliseconds per generated frame in 2X mode, compared with 1.26 ms for DLSS; in 4X mode, it took 0.61 ms versus 0.82 ms. At 4K, ANR used 214 MB of GPU memory, compared with 562 MB for DLSS, and its neural network weights occupied 0.17 MB. The company’s “up to 2.4X” claim comes from certain tests; the published RTX 5090 results show smaller processing-time differences. These figures measure the cost of generating additional frames, not equivalent gains in game FPS.

Apex’s comparisons and internal benchmarks report image-quality gains in difficult scenes involving particles, moving objects, water, vegetation, and disocclusion artifacts (visual errors where newly exposed areas are reconstructed). Its tests favored ANR over DLSS Frame Generation on CGVQM-5, ColorVideoVDP, and PSNR metrics from 2X to 6X frame generation. The company also released video showcases for The Last of Us Part II Remastered and Hogwarts Legacy, but independent testing is still needed.

Gamers can download the beta from GitHub and install it with OptiScaler, using a package suited to whether the game already has OptiScaler. The FG V.B14 release, published on November 12, fixes a Windows compatibility issue that kept some games from launching with older Visual C++ runtimes. Apex also released nr-record, a utility for capturing frame-generation inputs and replaying them through competing implementations for direct image-quality comparisons. ANR is not yet open-source; developers say they plan to release its source code later. As an experimental beta, it may have compatibility issues and visual artifacts.

Apex Compute introduced Apex Neural Rendering (ANR), a cross-platform neural rendering suite that the company says can generate frames up to 2.4X faster than NVIDIA DLSS while using 62% less GPU memory. Its Frame Generation component is available in beta; Super Resolution and Ray Reconstruction remain in development. The beta supports Windows 10/11, Linux, and SteamOS through Proton. A separate native Metal test kit supports Apple silicon on macOS 26 and newer.

ANR supports NVIDIA GeForce GTX 16 and RTX 20/30/40/50 Series, AMD Radeon RX 5000/6000/7000/9000 Series and Steam Deck, plus Intel Arc A-Series and B-Series. In Apex Compute’s internal RTX 5090 tests at 3840×2160, ANR took 1.01 milliseconds per generated frame in 2X mode, compared with 1.26 ms for DLSS; in 4X mode, it took 0.61 ms versus 0.82 ms. At 4K, ANR used 214 MB of GPU memory, compared with 562 MB for DLSS, and its neural network weights occupied 0.17 MB. The company’s “up to 2.4X” claim comes from certain tests; the published RTX 5090 results show smaller processing-time differences. These figures measure the cost of generating additional frames, not equivalent gains in game FPS.

Apex’s comparisons and internal benchmarks report image-quality gains in difficult scenes involving particles, moving objects, water, vegetation, and disocclusion artifacts (visual errors where newly exposed areas are reconstructed). Its tests favored ANR over DLSS Frame Generation on CGVQM-5, ColorVideoVDP, and PSNR metrics from 2X to 6X frame generation. The company also released video showcases for The Last of Us Part II Remastered and Hogwarts Legacy, but independent testing is still needed.

Gamers can download the beta from GitHub and install it with OptiScaler, using a package suited to whether the game already has OptiScaler. The FG V.B14 release, published on November 12, fixes a Windows compatibility issue that kept some games from launching with older Visual C++ runtimes. Apex also released nr-record, a utility for capturing frame-generation inputs and replaying them through competing implementations for direct image-quality comparisons. ANR is not yet open-source; developers say they plan to release its source code later. As an experimental beta, it may have compatibility issues and visual artifacts.