Osegi
Hardware·2 min read

Apex Neural Rendering: a DLSS rival that runs on any GPU

Apex Compute says its Apex Neural Rendering toolkit beats Nvidia's DLSS Frame Generation on speed and memory, and runs on GeForce, Radeon, Arc and Apple chips.

A graphics card on a test bench next to a screen showing a side-by-side comparison of upscaled game frames.
Image: 3DNews - Daily Digital Digest
SUMODialoediviewergiftdetect

Apex Neural Rendering (ANR) is a new DLSS alternative from Apex Compute, and the pitch is simple: similar or better image quality, less overhead, and no need for a recent Nvidia card. Only one piece is out so far, and it is in beta.

ANR is a toolkit with three parts: ANR Super-Resolution (an upscaler), ANR Frame Generation (which invents extra frames to smooth motion) and ANR Ray Reconstruction (which cleans up ray-traced lighting). Frame generation is available now. The upscaling and ray reconstruction models are still being trained and will come later.

How does ANR compare with DLSS Frame Generation?

The numbers come from the company. On a GeForce RTX 5090 at 4K with 2x scaling, ANR FG takes 1.01 ms per generated frame, against 1.26 ms for DLSS Frame Generation. It also uses 214 MB of video memory, where DLSS needs 562 MB. The neural network weights are just 0.17 MB. The test does not say which DLSS version was used, though TechPowerUp assumes it was the latest.

Apex Compute also published comparisons from Hogwarts Legacy and The Last of Us Part II Remastered. It claims ANR produces more faithful frames than DLSS, and in some cases better ones. These are the company's own examples, not independent tests.

Which graphics cards and systems does ANR support?

The beta is an injector built on OptiScaler, and it works in any game that supports Nvidia DLSS Frame Generation. Supported hardware includes:

  • Nvidia GeForce RTX 20 series and newer, plus GTX 16 series
  • AMD Radeon RX 5000 series and newer, and the Steam Deck
  • Intel Arc A and B series
  • Apple devices with M1 or newer chips

Tested operating systems are Windows 10, Windows 11, Linux, SteamOS with Proton, and macOS 26 or later.

Apex Compute also plans to release the whole ANR framework, all components included, as open source on GitHub. No date was given.

Apps worth a look

SUMO: the number puzzle

Tap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.

Tags:apex computeapex neural renderingdlssframe generationoptiscalergpu

More from this day

See all →

More in: Hardware

See all →