Photogrammetry vs Gaussian Splatting: The Real Difference and Hardware requirement for 3D workflow.

December 27, 2025

1. Photogrammetry

Photogrammetry is the industry standard for structure. It turns photos into a measurable, solid 3D object.

The Nitty-Gritty: How it Works It uses a process that thinks like a surveyor:

Finding Matches: The software looks at your photos and finds "key points"—corners or edges that look the same in every shot.
Connecting the Dots: It calculates exactly where those points are in 3D space.
The Mesh (The Main Event): It connects those dots to create a solid skin made of triangles.

Why we use it: If you need to measure a wall, print a statue, or put a character in a game that creates a shadow or blocks a path, you need this solid shape.

2. Gaussian Splatting (3DGS)

3D Gaussian Splatting example

It solves the hard problems: smoke, glass, hair, and shiny reflections.

What makes it different? Instead of a hard shell, 3DGS sees the world as a cloud of millions of "blobs" (Gaussians). Each blob has special superpowers that normal 3D points don’t have:

Covariance matrix in Gaussian Splatting

Shape Control (The "Splat"): Standard points are just dots. A Gaussian Splat can stretch. It can flatten out to look like a wall or stretch thin to look like a wire. This lets it paint complex shapes easily.

Spherical harmonics illustration

Smart Color (Spherical Harmonics): This is the big breakthrough. In a normal 3D model, a red car is just red. In 3DGS, the "splat" knows to change color based on where you are standing. As you move, the sunlight glints off the hood, or the velvet fabric changes shade. It captures the light, not just the surface.
Self-Correction: The system is smart. During training, if it sees a blurry area, it adds more splats automatically. If it sees empty space, it removes them. It optimizes itself to look perfect.

The Hardware: Fuelling the Engine

Both of these tools need power, but they need different kinds. Photogrammetry likes fast processors. Gaussian Splatting is hungry for Video Memory (VRAM) to manage all those millions of blobs.

To run this smoothly, you need the right setup:

For the Solo Creator (Scan & Train)

If you are scanning single objects or learning the ropes, you need speed for the initial setup.

The Choice: Pro Maven GS (Single GPU Desktop). This gives you the fast processor speed to align your photos in photogrammetry, plus a dedicated GPU to train your Gaussian Splats.

For the Pro Studio (Double the Speed)

In a studio, waiting is the enemy. You need to do two things at once.

The Choice: Pro Maven GT (Dual GPU Desktop). With two GPUs, you can multitask. Use one card to build a Photogrammetry mesh and the other to train a Splat of the same scene at the same time.

For the Heavy Lifters (City-Scale & Enterprise)

When you are mapping an entire city block or a massive factory, normal computers run out of memory. You need massive capacity.

The Workhorse: Pro Maestro GQ (4-GPU Server).
- Capacity: We load this with Pro 6000 or H200 cards so you have enough memory to handle millions of points without crashing.
- Speed: Or go with the 4x 5090 version for pure, raw speed.
The Research Lab: Pro Maestro GE (8-GPU Server). If you need to test 8 different lighting setups at once, this is the machine. It lets you run massive experiments in parallel using 5090s or Pro 6000s.
The Ultimate: Pro Maestro GD (10-GPU Server). This is as big as it gets. With 10 GPUs (Pro 6000 or H200), this server is built for the largest datasets in the world. It delivers the maximum power possible for city-scale 3D projects.