How long does it take to create a Gaussian Splat?

Gaussian Splatting processing time depends on the tool, input size, and available hardware. Here is a breakdown by platform:

Professional Processing (DJI Terra V5.0+)

• Per block: DJI publishes that each reconstruction block takes no more than 40 minutes (roughly 300 photos per block for a DJI M4E terrain-follow capture at 80% forward / 70% side overlap, on a 64 GB Windows machine)
• Large sites: processed as multiple blocks, so wall-clock time scales with the block count
• Maximum capacity: DJI publishes 12,800 photos per standalone task at 32 GB RAM and 25,600 at 64 GB (192,000 / 384,000 in cluster mode)

Consumer / Cloud Processing

Hardware Requirements

For local GS processing, GPU power determines speed:

• Minimum: 8GB VRAM (e.g., RTX 3060) — handles small scenes
• Recommended: 16–24GB VRAM (e.g., RTX 4080/4090) — handles large scenes efficiently
• System RAM: 32GB minimum for large datasets, 128GB for city-scale reconstructions
• Storage: SSD recommended — GS generates large intermediate files

Training vs Rendering

It is important to distinguish:

• Training time (creating the GS model): Minutes to hours — this is the computationally expensive step
• Rendering time (viewing the result): Real-time at 60+ FPS on any modern GPU — this is GS's key advantage over NeRF

Comparison to Other Methods

For a professional Gaussian Splatting project, THE FUTURE 3D handles all processing — you receive finished deliverables without needing GPU hardware. Get a quote →