1.3 Computer System Requirements

The Short Version

One workstation can run both LixelStudio and LCC Studio. Buy against the work you actually take on, because the two applications have different bars and LCC Studio sets the higher one.

What Each Part Buys You

• RAM: how long a scan you can process at all. This is the wall most people hit first, and it is a wall rather than a slope.
• GPU: how fast a reconstruction finishes, and whether the software runs at all. The brand and generation are pass or fail; the model is speed.
• SSD: processing speed, plus the working space a job needs to finish without failing partway.
• CPU: the most leverage on Map Fusion and Aerial-Ground Fusion. Modest effect on single-model work.

Three Builds

These map to real work. Pick the row that matches the longest scan and the heaviest reconstruction you expect to run, not the one you run today.

Published Requirements, Side by Side

These are the manufacturer's published figures. LCC Studio, which produces the 3D Gaussian Splat models, is the more demanding of the two on every line.

LixelStudio (Point Clouds: L2 Pro and K2)

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

What Will Not Work at Any Price

These are pass or fail. No amount of money spent elsewhere in the build compensates, and every one of them has cost somebody a returned machine.

Graphics

• NVIDIA only. AMD and Intel graphics cards are not compatible with either application, and no driver update changes that. AMD processors, by contrast, are fully supported and appear in the manufacturer's own recommended builds.
• RTX 20 series or above. Older 10-series cards are to be avoided regardless of memory size.
• CUDA 11.6 or above, and NVIDIA driver version 580 or above. Check both before buying a used or older card. A card that clears the model name and fails the CUDA level still will not work.
• One GPU only. Multi-GPU parallel processing is not supported; the software selects a single GPU automatically. A second card adds nothing to processing.
• Watch integrated graphics on machines that have both. Windows may default to the integrated chip rather than the discrete card, which looks like a performance problem and is a settings problem.

Operating System and Platform

• Windows 10 or 11, 64-bit, only. The standard edition does not support virtual machines, Windows Server, Linux in any distribution, or macOS. A Mac with Windows in a virtual machine is not a supported configuration.

Storage

• Process from a local internal disk. External drives are a documented cause of post-processing crashes. USB flash drives and network or LAN paths carry the same problem. Copy the project to an internal SSD first.

Laptops

• Use a desktop for large dataset processing. Laptops are prone to thermal throttling under sustained heavy loads, which can cause crashes. A reconstruction runs for hours at full load, which is exactly the condition a laptop chassis is worst at. A laptop meeting the published figures will handle short scans; specify a desktop for production work.

Memory, Storage, and Where They Bite

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

Both applications hold the working reconstruction in memory, so memory sets a hard wall rather than a smooth slope. Below the wall, more memory does nothing for speed. At the wall, it is the difference between a finished model and a failed job.

Two consequences follow. Scans longer than your memory supports must be split into segments and joined with Map Fusion, so the memory decision and the field segmentation plan are the same decision. And on a machine at its limit, dropping reconstruction efficiency to Standard reduces memory demand and can carry a job that would otherwise fail.

Storage: Size the Free Space, Not Just the Drive

A drive large enough to hold the data is not large enough to process it. Reserve free working space before starting a job, or it can fail partway through.

The 1 mm enhancement multiplier is not a typo. A 2 GB point cloud needs close to a terabyte of free space to enhance at 1 mm. Check the figure before committing to that deliverable, and keep project files on a dedicated SSD separate from the Windows drive and the install folder. Alpine's field estimate for raw scan data is 60 to 80 GB per hour of scanning, which is what to budget capacity against.

Map Fusion and Aerial-Ground Fusion

These workloads align high-density multi-model data and demand more than a single-model job. The published configuration:

• CPU: 16-core or better high-performance desktop processor, such as an AMD Ryzen 9 9950X.
• RAM: 64 GB DDR5 minimum. 128 GB for larger-scale data, and for projects with total scan length of 150 minutes or more at high quality.
• GPU: NVIDIA RTX 3090 minimum. RTX 4090 or 4090D for best performance.

Confirm resources are sufficient before starting a fusion task rather than after it fails. If memory is short for a large project, select Standard reconstruction efficiency to reduce demand and keep the job stable.

What Affects Processing Speed

There is no definitive reconstruction time estimate, because it depends on the data. The published planning figure is a 1:20 ratio under the recommended configuration at Standard reconstruction efficiency: one minute of capture data takes roughly twenty minutes of processing. A 15-minute capture is therefore most of a working day of compute. Quote schedules against that, not against a best case.

• Reconstruction efficiency setting. The biggest control you hold on any single job, and it costs nothing. Higher quality extends time; Standard reduces both time and memory demand.
• GPU class. A high-end card is faster than a mid-range one, and the gap widens on high-resolution data. A mid-range card is stable, just slower under heavy load. This is the main hardware speed lever.
• Storage medium. An SSD is significantly faster than a spinning drive, and processing from an internal disk avoids the crashes external drives cause. Fix this before spending on the GPU.
• Scan size and job type. Processing scales with capture length, and fusion is heavier than a single model.
• Other work on the machine. Keep the workstation dedicated while a job runs. Set the Windows power plan to High Performance so sleep or power saving does not interrupt processing.

Point Cloud Enhancement Costs Time as Well as Disk

Measured on a machine with an RTX 4090 and 64 GB of memory: an indoor small-area project runs about 1:2 at 5 mm and about 1:8 at 1 mm. An outdoor large-area project runs about 1:5 at 5 mm and about 1:41 at 1 mm. Combined with the 465 times disk requirement, 1 mm enhancement on a large outdoor dataset is a scheduling decision rather than a checkbox.

Where an Upgrade Changes Outcomes

Above the Standard build, one upgrade changes what you can do and the rest buy speed.

• Memory from 64 GB to 128 GB lifts the capture-length wall from about 30 minutes to about 60, and is the published requirement for larger fusion projects. This is the only upgrade on the list that changes which jobs you can accept.
• A faster GPU or CPU past the recommended bar buys shorter processing times on the same work. Worth it if compute time is your bottleneck; not worth it if capability is.
• Two free wins that need no purchase: drop the reconstruction efficiency setting when quality allows, and keep project data on an internal SSD rather than an external drive.

Manufacturer specifications change between software releases. Confirm current requirements at the XGRIDS documentation site, or ask Alpine Reality Capture, before committing to a hardware purchase.