1.3 Computer System Requirements

The Short Version

One workstation can run both LixelStudio and LCC Studio. The build below clears the recommended bar for both applications. If you are buying, the single upgrade that changes outcomes is RAM. More RAM does not make a given scan finish faster. It raises the longest scan the software can process at all.

What Each Part Buys You

• RAM: how long a scan you can process at all. This is the wall most people hit first.
• VRAM: which LCC Studio features run, and at what quality.
• GPU class: how fast a reconstruction finishes.
• SSD: processing speed, plus the working space a job needs to complete without failing.
• CPU: the most help on LCC Studio Map Fusion. Modest effect on single-model work.

Official Requirements, Side by Side

These are the published minimums and recommendations. Running below the minimum causes mid-run failures, not slow processing. The software can appear to work and then crash, losing the work done up to that point.

LixelStudio v4.0 (Point Clouds: L2 Pro, K2)

LCC Studio v1.13 (3D Gaussian Splats: L2 Pro, K2, PortalCam)

What Each Component Actually Controls

RAM Is the Scan-Length Ceiling, Not a Speed Dial

LixelStudio holds the entire scan trajectory in memory during SLAM optimization, using roughly 2 GB of RAM per minute of scan. LCC Studio behaves the same way at the project level. RAM sets a hard wall, not a smooth slope. Below the wall, more RAM does nothing for speed. At the wall, it is the difference between a finished model and a failed job.

Two practical consequences. First, the jump from 64 GB to 128 GB roughly doubles the longest scan you can process, so it is the highest-value upgrade for anyone running scans over 30 minutes. Second, scans longer than your RAM supports must be split into segments and joined with Map Fusion. LixelStudio 1 mm point cloud enhancement also requires 128 GB.

VRAM Is the Feature and Quality Gate

In LCC Studio, video memory decides what you can build and which features run, more than how fast a job finishes. It behaves in cliffs, not a slope.

• Above 8 GB unlocks two features. HD Enhancement and Spatial Recognition require more than 8 GB of VRAM. Below that they will not run at all.
• Model size ceiling. Max Gaussian Points must stay within VRAM capacity, typically 25M or under in Single Model mode. Set it too high and you get VRAM shortage, performance degradation, and reduced model quality.
• Quality tier. Slow reconstruction quality increases VRAM use and processing time in exchange for higher quality. Enough VRAM lets you run Slow; too little forces Standard or Fast.

GPU Class Is Where Reconstruction Speed Comes From

A high-end card such as the RTX 4090 reconstructs faster than a mid-range card such as the RTX 3060, and the gap widens on large or high-resolution point clouds. Both applications require an NVIDIA GPU with CUDA. AMD GPUs are not supported, and no driver update changes that. AMD CPUs, by contrast, are fully supported.

CPU Matters Most for Map Fusion

For single-model work, the CPU has the least leverage of the four components. It matters most for LCC Studio Map Fusion and Aerial-Ground Fusion, which want a 16-core class processor. XGRIDS recommends the AMD Ryzen 9 9950X or equivalent for that work.

SSD Is the Cheapest Real Speedup

Putting the project path and data directory on an SSD measurably improves generation and processing speed. Two storage rules:

• Reserve at least 2x the capture data size in the LCC Studio data directory before starting, or the job can fail partway through.
• Keep project files on a dedicated SSD, separate from the Windows drive and from the install folder. Raw scan data runs about 60 to 80 GB per hour, so budget capacity before you scan.

What Affects Processing Speed

Four things move processing speed, and the most powerful one is free: the quality setting you pick per job. RAM and VRAM are not primarily speed levers. When they are short, they slow you down indirectly by forcing fallback modes.

• Quality setting (Fast, Standard, Slow). The biggest control you hold per job. Slow extends time for higher quality. Standard runs about 20 minutes of processing per 1 minute of capture. Overall processing lands around 1:20 to 1:30, so a 5-minute scan takes roughly 100 to 150 minutes.
• GPU class. A higher-end card finishes faster, especially on large data. This is the main hardware speed lever.
• Storage medium. An SSD is significantly faster than a spinning drive. Fix this before spending on the GPU.
• Scan size and type. Processing scales with capture length. Map Fusion and Aerial-Ground Fusion are heavier than a single model.

Three indirect slowdowns tie back to the memory points above. Insufficient RAM forces Low-Memory Mode, which writes intermediate data to disk and extends reconstruction time in exchange for stability. Insufficient VRAM causes performance degradation. Running other GPU-heavy tasks during a job competes for VRAM, so keep the workstation dedicated while a job runs. Finally, SLAM Auto mode retries in Robust mode after a failure, and each retry costs time, so picking the SLAM mode that matches the environment up front avoids wasted passes.

Heavier Requirements for Map Fusion and Aerial-Ground Fusion

These LCC Studio workloads align high-density multi-model data and demand more than a single-model job. Per the LCC Studio v1.13 manual:

• CPU: AMD Ryzen 9 9950X or an equivalent 16-core high-performance desktop processor.
• RAM: 64 GB DDR5 base. 96 to 128 GB for projects with total scan duration of 150 minutes or more.
• GPU: NVIDIA RTX 3090 minimum. RTX 4090 or 4090D recommended.

If RAM is insufficient for a large project, use Standard quality to reduce memory demand and keep the job stable.

Where an Upgrade Changes Outcomes

Above the recommended build, two upgrades change what you can do. Everything else is incremental.

• RAM from 64 GB to 128 GB lifts the scan-length wall from about 30 minutes to about 60 minutes and is required for 1 mm point cloud enhancement and large Map Fusion projects.
• GPU VRAM from 8 GB to 12 or 16 GB unlocks LCC Studio HD Enhancement and Spatial Recognition and adds headroom for higher point counts and Slow quality.

A faster CPU or a higher GPU clock past the recommended bar buys incremental speed, not new capability. The two free wins that need no purchase: drop the quality setting from Slow to Standard, and keep project data on an SSD.