System Requirements
Minimum Specifications
Section titled “Minimum Specifications”Onset Engine is designed to be accessible while still delivering professional-grade results. To run the application, your system must meet these minimum requirements:
- OS: Windows 10 64-bit or higher (Onset Engine is currently Windows-only - there is no macOS or Linux version yet. Get notified when the Mac version is ready)
- CPU: Any x86 CPU from 2020 or newer (Intel Core i5/i7 10th Gen+, AMD Ryzen 3000 series+)
- RAM: 8 GB
- Storage: 5 GB of free disk space (for the application and bundled AI models)
Recommended Specifications
Section titled “Recommended Specifications”For the best experience, especially when working with 4K footage or large clip libraries, we recommend the following:
- OS: Windows 11
- CPU: Intel Core i7/i9 12th Gen+ or AMD Ryzen 5000 series+
- GPU: NVIDIA RTX 3070 or better (with at least 8GB VRAM). RTX 50-series cards are fully supported.
- RAM: 16 GB or more
- Storage: NVMe SSD for fast media access and caching
Hardware Compatibility
Section titled “Hardware Compatibility”Onset Engine adapts to your hardware automatically. Here’s what works and how well:
| Capability | NVIDIA (CUDA) | AMD Radeon | Intel (CPU-only) |
|---|---|---|---|
| AI Ingest (CLIP) | ✅ GPU-accelerated (fastest) | ⚙️ CPU fallback (works, slower) | ⚙️ CPU fallback (works, slower) |
| Video Encoding | ✅ NVENC (h264/HEVC) | ✅ AMF (h264/HEVC) | ⚙️ libx264/libx265 (CPU) |
| Video Decoding | ✅ Hardware-accelerated | ⚙️ CPU | ⚙️ CPU |
| VFX Rendering | ✅ CPU + multithreaded | ✅ CPU + multithreaded | ✅ CPU + multithreaded |
| Face Detection | ✅ CPU | ✅ CPU | ✅ CPU |
| Depth Effects | ✅ GPU or CPU | ⚙️ CPU | ⚙️ CPU |
| Beat Detection | ✅ GPU-accelerated (CUDA) | ⚙️ CPU | ⚙️ CPU |
Key: ✅ = fully supported | ⚙️ = works via fallback
Which NVIDIA GPUs get accelerated AI analysis?
Section titled “Which NVIDIA GPUs get accelerated AI analysis?”AI analysis (CLIP ingest, beat detection, depth effects) runs on NVIDIA GPUs from the Turing generation onward. Encoding (NVENC) and decoding are unaffected by this and work on a wider range of cards.
| GPU generation | Example cards | AI analysis |
|---|---|---|
| Blackwell (RTX 50-series) | RTX 5070 Ti, 5080, 5090 | ✅ GPU-accelerated |
| Ada Lovelace (RTX 40-series) | RTX 4060 – 4090 | ✅ GPU-accelerated |
| Ampere (RTX 30-series) | RTX 3060 – 3090 | ✅ GPU-accelerated |
| Turing (GTX 16 / RTX 20-series) | GTX 1660, RTX 2060 – 2080 | ✅ GPU-accelerated |
| Pascal and older (GTX 10-series and earlier) | GTX 1060, 1080, GTX 900-series | ⚙️ CPU fallback |
Cards older than Turing still run Onset Engine completely - ingest simply runs on the CPU, which is slower but produces identical results. NVIDIA retired these architectures in CUDA 13, and the AI libraries Onset Engine builds on no longer ship kernels for them.
Note: AMD users get full hardware encoding (AMF) for fast exports. The only difference vs. NVIDIA is that AI analysis runs on the CPU instead of the GPU. With the default Standard analysis, importing footage takes only slightly longer; with High Quality analysis it takes several times longer (see the timings below). Render quality and speed are not affected.
The Role of the GPU
Section titled “The Role of the GPU”A dedicated GPU is completely optional but highly recommended.
Here is how Onset Engine utilizes your hardware:
- Ingest & AI Analysis (CLIP): When you ingest footage, the engine extracts frames and analyzes them using AI (CLIP embeddings, scene detection, motion analysis). An NVIDIA GPU (via CUDA) runs the AI part of this on the GPU. Without one, the engine automatically falls back to CPU inference, which gives identical results. How much slower that is depends on your analysis settings - see the timings below.
- Hardware Encoding (NVENC): For exporting video, Onset Engine supports hardware-accelerated encoding (like
h264_nvencandhevc_nvenc) on compatible NVIDIA GPUs, drastically reducing final render times. - Rendering & VFX: The actual video rendering and visual effects pipeline (via MoviePy and our custom VFX core) relies heavily on the CPU. The engine uses multithreading (
max(4, os.cpu_count()-2)) to process frame transforms in parallel. Therefore, rendering works great on a fast CPU, even without a powerful GPU.
Measured Ingest Times
Section titled “Measured Ingest Times”Time to import 10 minutes of 1080p H.264 footage (four files), measured on an Intel Core i5-12400F (6 cores) with an NVIDIA RTX 3060 Ti. The “CPU only” column is the same PC with the GPU disabled for AI analysis, which is how an AMD or Intel-graphics system runs.
| Analysis settings | With NVIDIA GPU | CPU only (AMD / Intel) |
|---|---|---|
| Standard (default), 1 frame/s | ~45 s | ~57 s (1.3× longer) |
Standard, sample_fps: 4.0 | ~47 s | ~2 min (2.7× longer) |
| High Quality, 1 frame/s | ~48 s | ~4.5 min (5.7× longer) |
With an NVIDIA GPU, reading and decoding the video is what limits speed, so the analysis settings barely matter. Without one, the AI model becomes the bottleneck: the heavier the settings, the bigger the gap. Faster CPUs close some of that gap; high-resolution (4K) or HEVC footage takes longer on any system.
Render times depend mostly on the style preset and output resolution - see Render Timing Benchmarks.
Codecs and Formats
Section titled “Codecs and Formats”Onset Engine supports all major video containers (MP4, MOV, MKV, AVI, WMV, WebM) and exports using H.264 or HEVC encoding. FFmpeg and MPV are bundled with the installer - no additional software is needed.