Minimum (functional, slow)
The viewer and annotation tools work without a GPU, but the vessel detection pipeline runs entirely on CPU and is materially slower. This is fine for reviewing pre-computed segmentations, not for running new cases.
System requirements
What the Spwig Medical Viewer needs to run — and what makes it fast.
Spwig Medical Viewer can run on CPU alone, but realistic vessel detection pipelines move from minutes to hours without a GPU. The numbers below assume a 512×512×400 CTA volume — a typical thin-slice cerebrovascular study.
Recommended (fast, fluid)
A modern NVIDIA GPU with 16 GB VRAM puts the full pipeline under a minute and keeps the 3D annotation studio responsive.
| Component | Minimum | Recommended | Tested |
|---|---|---|---|
| GPU | None (CPU only) | NVIDIA, 16 GB VRAM | RTX 4090 Laptop (16 GB) |
| NVIDIA driver | — | ≥ 535 (Linux) / ≥ 537 (Windows) | 580.159.03 |
| CUDA | — | 12.x compatible | 12.x |
| CPU | Modern 8-core x86_64 | 12+ cores | i9-13900HX (24 cores) |
| System RAM | 32 GB | 64 GB | 64 GB |
| Disk | ~15 GB free | NVMe SSD, ~30 GB free | NVMe SSD |
| OS | Windows 10+, Linux x86_64 | Windows 11, Ubuntu 22.04+ | Ubuntu 24.04 LTS |
Approximate performance
Runtime for the full vessel detection + bone separation pipeline on a 512×512×400 CTA volume.
| Hardware | Pipeline runtime |
|---|---|
| RTX 4090 Laptop, 16 GB VRAM (tested) | 30–60 seconds |
| RTX 3060 / 3070 Ti, 8–12 GB VRAM | 1–3 minutes (smaller batches) |
| 16-core CPU, no GPU | 20–60 minutes |
Tested hardware
This is the laptop we run development and benchmarking on. Performance numbers on this site ("~30–60 seconds per scan") refer to this configuration.
- cpu
- Intel Core i9-13900HX (24 cores, 32 threads)
- ram
- 64 GB DDR5
- gpu
- NVIDIA RTX 4090 Laptop GPU, 16 GB VRAM
- driver
- NVIDIA driver 580.159.03 (CUDA 12.x)
- os
- Ubuntu 24.04 LTS
- storage
- NVMe SSD
What about macOS?
Apple Silicon Macs don't have NVIDIA GPUs and Spwig Medical Viewer's pipeline is CUDA-based. We don't ship a Mac build today. The viewer (without the pipeline) could theoretically run on Mac in CPU-only mode for reviewing pre-computed segmentations — let us know if that's useful to you and we'll prioritise it.