Apex Neural Rendering: a cross-vendor frame-generation beta installed via OptiScaler, plus nr-record, a tool to record and compare frame-generation and upscaler output.
- Language
- C++
- Stars
- 1.8K
- Try it
Download anr-fg-full (or anr-fg-addon for games with OptiScaler) from Releases, then run install.cmd on Windows or sh install.sh "<game folder>" on Linux/Proton
A company called Apex Compute has put a free, cross-vendor frame-generation beta on GitHub that it says outperforms Nvidia's DLSS. The apex-compute/ANR repository was created on November 11 and had 1,841 stars and 29 forks by late November 14, according to GitHub's search API. Apex announced Apex Neural Rendering (ANR) on X the same day, pitching it as a combined super-resolution, ray-reconstruction and frame-generation suite.
Why it matters: Nvidia's own frame generation is tied to its newer RTX cards (ANR's README notes that replaying DLSS-G needs an RTX 40 or 50 series GPU). ANR's beta lists support back to Nvidia's GTX 16 series, AMD's RX 5000 series and the Steam Deck, and Intel Arc. If its claims hold up, it would give owners of older or non-Nvidia GPUs a new frame-generation option. That's a big if, and the repo is clear about what is still missing.
What's actually in the repo
Per the README:
| Component | Status |
|---|---|
| ANR FG (frame generation) | Beta, downloadable from Releases |
| ANR SR (super resolution / upscaling) | "Beta soon" |
| ANR RR (ray reconstruction / ray-tracing denoiser) | "Beta soon" |
| ANR source code | "Coming soon" |
| nr-record (capture and compare tool) | Source included |
So the part people will actually run, frame generation, ships as binaries only for now. GitHub shows no licence on the repo, which is why GamingOnLinux declined to call the project open source. Wccftech reports the developers have promised to release the source later.
How it installs
ANR FG piggybacks on OptiScaler, the community tool that swaps upscalers and frame generators in games. The README offers two packages: anr-fg-full for games without OptiScaler and anr-fg-addon for games that already use it. You close the game, extract the package and point the installer at the folder containing the game's .exe:
# Windows
install.cmd
# Linux / Steam Deck (Proton)
sh install.sh "<game folder>"
# then set Steam launch options to:
# WINEDLLOVERRIDES="dxgi=n,b" %command% (winmm=n,b for Vulkan games)
Then you turn on frame generation in the game, picking the Nvidia option where there's a choice. Uninstall scripts restore the original files. It runs on Windows 10 and 11, and on Linux and SteamOS through Proton. Wccftech also mentions a separate native Metal test kit for Apple silicon on macOS 26 and later.
The most interesting piece for sceptics may be nr-record. It losslessly records the inputs a game feeds its frame generator, then replays the same recording through ANR FG, a game's own DLSS-G library, or DLSS, FSR and XeSS upscalers, and produces synced side-by-side videos. That gives reviewers a way to check Apex's quality claims on identical frames instead of relying on its screenshots. DLSS-G replays need an RTX 40 or 50 series card.
The claims, and why to be cautious
Apex's own benchmarks, as reported by Wccftech and Windows Central, were run on an RTX 5090 at 4K:
| Metric | ANR FG | DLSS FG |
|---|---|---|
| Cost per generated frame, 2X mode | 1.01 ms | 1.26 ms |
| Cost per generated frame, 4X mode | 0.61 ms | 0.82 ms |
| VRAM at 4K | 214 MB | 562 MB |
Those are real gains on paper, but they aren't the "up to 2.4×" in Apex's headline. Wccftech notes the published figures are more modest and measure frame-generation overhead, not game FPS. Windows Central's writer suspects the 2.4× figure applies to a feature that isn't released yet. Both outlets also note the tests used a single flagship card, with no results yet on the older GPUs ANR is pitched at. Apex reports image-quality scores (CGVQM-5, ColorVideoVDP, PSNR) favouring ANR, but those come from the company as well. None of the outlets we read has tested it independently, and neither have we.
There is also little public information about Apex Compute itself. GamingOnLinux and Windows Central both say they couldn't find background on the company, and its GitHub organisation page gives no bio.
Who should try it
- Tinkerers with GTX 16, RX 5000/6000 or Steam Deck hardware who already use OptiScaler: it's a low-cost experiment, and the README asks for performance results and bug reports as GitHub issues. Expect artifacts and compatibility problems; Wccftech describes it as an experimental beta.
- Reviewers and modders: nr-record is worth a look on its own for like-for-like comparisons.
- Everyone else: wait for the upscaler, the source release and a licence, and for independent tests on mid-range cards. The README points to the FG-V.B15 release for nr-record 1.1.
Sources
- GitHub — apex-compute/ANR (README) github.com
- Wccftech — Apex Neural Rendering Challenges NVIDIA DLSS With Up To 2.4X Faster Frame Generation, 62% Lower VRAM Usage wccftech.com
- GamingOnLinux — Apex Compute revealed Apex Neural Rendering (ANR), their cross-platform answer to NVIDIA DLSS gamingonlinux.com
- Windows Central — Apex Compute's free ANR might be the DLSS alternative for older GPUs windowscentral.com