GPU servers for rendering and generation
Rendering and image or video generation split nicely across cards, which makes multi-GPU consumer and workstation machines very good value.
What to pick
- Blender Cycles, Redshift, Octane: 4× or 8× RTX 5090 / 4090 — every card renders its own frames.
- Heavy scenes that don't fit in 24–32 GB: RTX 6000 Ada (48 GB) or RTX PRO 6000 (96 GB).
- Image and video generation (ComfyUI, diffusion models): RTX 5090 for speed, RTX PRO 6000 or H100 for the largest video models.
How it works
These are Linux machines that you control over SSH — there is no remote desktop. Upload your project, run the renderer from the command line, and download the frames. Blender, for example, renders headless with blender -b scene.blend -a.
For browser tools like ComfyUI, start them on the server and open them through an SSH tunnel.
Files
Each machine has 100 GB of NVMe disk. Copy results off before the period ends — the server and its disk are deleted when it expires.