10.7 Device Specifications
Side-by-side comparison of the L3, L2 Pro, K2, and PortalCam, complete specification tables for each device, and the processing workstation requirements for both software pipelines.
Choosing a Device
Output type, georeferencing, and mount options are the three that most often decide whether a device can do a job at all. The full specification tables follow, starting with the L3. The L3 against the L2 Pro in more depth is on the L3 buying guide.
The two duration rows are different facts that happen to share their numbers. Maximum single capture is a firmware limit: recording stops there whatever the power situation. Runtime per battery is how long a charge lasts. Neither is a scan target, and on most workstations the reconstruction memory in Section 6 will bind long before either.
The L2 Pro, K2, and PortalCam cannot resume a scan from a breakpoint. If a scan stops for any reason, that recording is closed and the next scan is a new segment. Recovery is Map Fusion, not resumption, and Map Fusion needs overlap you must have already captured. Plan segment boundaries before you start rather than discovering one mid-scan.
Lixel L3
The L3 is XGRIDS' high-accuracy spatial scanner. It ships in two configurations, 16-channel / 120 m and 32-channel / 120 m, which differ in channel count and point rate and share every other specification below. Survey-grade RTK and a 4.3-inch display are built in, and one capture produces a true-color point cloud, a mesh, and an LCC 3DGS scene.
Conditions attached to the L3 accuracy figures. Relative accuracy is stated at 1 sigma, not RMSE, from a post-processed capture made without ground control points in a dedicated 5,382 sq ft (500 m²) indoor test environment, so it is the result of a different test from the L2 Pro and K2 relative figures below. Absolute accuracy, repeatability, real-time absolute accuracy and leveling with RTK fusion are measured under open sky, free of multipath, with optimal GNSS geometry and favorable atmospheric conditions. Real-time relative accuracy and leveling without RTK fusion are measured under laboratory conditions. Point cloud thickness is planar thickness within a 10 m (33 ft) walking path. A figure quoted without its condition is not the figure XGRIDS published. The method is in XGRIDS' Accuracy Verification White Paper.
Lixel L2 Pro
The L2 Pro ships in three configurations differentiated by LiDAR channel count and working range. The sensor, camera, accuracy, and physical specifications are identical across all three.
L2 Pro 16/120
Channels: 16
Working range: 0.5 m to 120 m (1.6 ft to 394 ft)
Scan speed: 320,000 pts/s
Entry configuration. Sufficient for most AEC use cases with building heights under 394 ft (120 m).
L2 Pro 32/120
Channels: 32
Working range: 0.5 m to 120 m (1.6 ft to 394 ft)
Scan speed: 640,000 pts/s
Higher point density than 16/120. Recommended configuration for most professional use. Best scan efficiency of the three models.
L2 Pro 32/300
Channels: 32
Working range: 0.5 m to 300 m (1.6 ft to 984 ft)
Scan speed: 640,000 pts/s
Extended range for tall structures and large outdoor environments. Better drift management than 120 m models at distance. Building height support up to approximately 590 ft (180 m).
Reading the output format rows across devices. XGRIDS publishes output at two different levels of detail. The L2 Pro sheet lists what the device itself writes (.las and .jpg); the L3 and K2 sheets list what the full pipeline produces, including mesh, 3DGS, 3D Tiles and USD. The L3, L2 Pro, and K2 all feed LixelStudio and LCC Studio, so all three reach the same export set. The L3 rows above and the K2 rows below are longer because their source documents are more complete, not because either device outputs more than the L2 Pro.
Conditions attached to the L2 Pro accuracy figures. Absolute accuracy assumes no continuous RTK or GCP disconnection longer than 328 ft (100 m). Relative accuracy applies between two points less than 328 ft (100 m) apart. Repeat accuracy assumes two RTK-enabled scans with no disconnection. Point cloud thickness is measured within 33 ft (10 m) of the walking path. A figure quoted without its condition is not the figure XGRIDS published.
Storage planning for L2 Pro. The 1 TB internal SSD holds approximately 12 to 16 hours of raw scan data at 60 to 80 GB per hour of scanning. Offload data to a processing workstation after each project. The SSD is not swappable; the only offload method is USB 3.1 Gen2 data transfer in USB mode or via the LixelGO app.
Lixel K2
The K2 is XGRIDS' second-generation handheld scanner. It replaces the previous handheld in current production with extended range, real-time true-color point cloud rendering, and built-in RTK. The K2 ships ready for georeferenced scanning without any external module. K2 is handheld only; it does not support backpack, vehicle, or drone mounts.
Guided Mode and RGB Mode are a pre-scan choice you cannot reverse. Selecting RGB Mode gives you real-time true colour but removes the Scan Assist System and the Scan Quality view, which are the only in-field checks on coverage density. Selecting Guided Mode keeps those checks but shows elevation or quality colouring rather than true colour. If nobody is reviewing coverage on site, use Guided Mode and check the Scan Quality view before leaving. Use RGB Mode when a client is watching the screen or when true-colour confirmation matters more than a coverage check.
K2 reconstruction modes in LCC Studio. The K2 supports single-scene reconstruction, Map Fusion with up to 10 sub-scenes per run, and Aerial-Ground Map Fusion. Single-scene length is bounded by the 90-minute device capture ceiling and by the memory available on the processing workstation; LCC Studio itself imposes no time limit. Aerial-Ground Map Fusion requires the ground scan data, the drone imagery, and the takeoff and landing point photos together, and reconstruction cannot start if any one of the three is missing. See Module 9 for the full processing workflow.
Two figures for K2 absolute accuracy, and both come from XGRIDS. The published specification sheet states 3 cm RMSE. The dealer New Product Introduction states 3 cm with a 5 cm maximum, on both the real-time and post-processed tables. The rows above carry the fuller dealer figure because it is the one that matters when a client holds you to a tolerance. Quote 3 cm as the specification and plan for up to 5 cm on demanding work. All absolute figures assume open sky, no multipath, good satellite geometry, and favourable atmospheric conditions.
Storage planning for K2. The 512 GB internal eMMC holds approximately 6 to 8 hours of raw scan data. Internal storage is not field-swappable. Offload data to a processing workstation after each project via USB 3.1 Gen2 or LixelGO.
PortalCam
The PortalCam pairs a LiDAR with 180 by 180 degree coverage and a four-camera array whose combined fisheye field exceeds a full sphere, optimized for close-range, high-fidelity indoor capture. It outputs 3D Gaussian Splat models through LCC Studio only. Point cloud output is not supported. It does not produce absolute georeferenced coordinates.
IP rating note. XGRIDS has not published an IP rating for the PortalCam. Treat the device as unrated. Do not expose it to rain, high humidity, or dusty environments, and do not assume it matches the other scanners, which are rated IP54.
Direct sunlight. The PortalCam operating temperature range applies in indirect sunlight only. Do not scan in direct sun for extended periods, and do not leave the device in direct sun between segments. Heat soak in an enclosed vehicle or on an exposed rooftop will exceed the rated range well before ambient air temperature does.
What the PortalCam geometry figures mean. The relative and repeat figures above describe internal geometric consistency within a reconstruction. They are not a measurement tolerance and they are not an absolute accuracy specification, because the PortalCam produces no georeferenced output. Do not quote them in a proposal or a scope of work. If a client needs a number they can hold you to, the job needs an L3, an L2 Pro, or a K2.
Segment planning across batteries. Three separate numbers bound a PortalCam job and they are easy to run together. The firmware stops a recording at 60 minutes. A battery lasts about 60 minutes of continuous scanning. XGRIDS capture guidance is 30 minutes or less per capture, which is the number that should drive your plan and is what makes at least two segments per battery achievable. Fusion adds its own two caps: 10 segments and 200 minutes combined. Reconstruction memory climbs with scan length on top of all of it, so a working segment length well below 30 minutes is the practical target. See 7.4 PortalCam Map Fusion.
Processing Computer Requirements
Both pipelines are Windows only and both require an NVIDIA GPU. The specifications below are the published requirements for each application. Buy or build against the workflow you actually run, because Map Fusion and Aerial-Ground Fusion demand more than single-scene reconstruction does.
LixelStudio
Point cloud processing.
LCC Studio
3D Gaussian Splatting reconstruction. LCC Studio Basic and LCC Studio Premium tiers. Current pricing is on the XGRIDS Store.
Map Fusion and Aerial-Ground Fusion
Multi-segment stitching and high-compute reconstruction need more than the recommended single-scene configuration. Running fusion below these thresholds produces out-of-memory failures and incomplete fusion outputs.
Memory and Scan Length
Memory is what limits a single reconstruction on a real workstation. Neither the battery nor the device capture ceiling is the operative constraint, and there is no software ceiling at all: LCC Studio imposes no time-based rejection and no built-in timer, so it will accept and attempt a scan far larger than the machine can finish.
The software predicts the failure before it happens, if you read the estimate. LCC Studio estimates memory demand from the uploaded data and displays it before the run. A red Estimated Memory reading means the job is too large for this machine, and the fix at that point is to drop Reconstruction Quality to Standard, add memory, or split the capture. Committing past a red estimate ends in an out-of-memory failure with nothing to recover. The error code and the workflow are on 9.2 Single Scene Processing.
Alpine working target: 12 to 15 minutes per segment. The table above is the ceiling, not the plan. A failed reconstruction on a 45-minute scene costs a day and cannot be partially recovered, and reconstruction time scales with scan length at roughly 20 minutes of processing per minute of capture. Planning at 12 to 15 minutes leaves room for unexpected coverage without approaching either limit. XGRIDS makes the same point from the field side: it does not recommend capturing a full-length dataset in one session on any hardware, because a single fault inside a long capture can cost the entire dataset.
Disk Headroom
Configurations that will not work, whatever the specification. Neither application runs on a virtual machine, on Windows Server, on Linux, or on macOS. Neither uses more than one GPU: multi-GPU parallel processing is not supported and the software selects a single GPU automatically, so a second card adds nothing to processing. AMD and Intel graphics cards are unsupported with no workaround, though AMD processors are fully supported and appear in the recommended configurations above. For LixelStudio the NVIDIA card must be NVIDIA RTX 20 series or above with CUDA 11.6 or later, running NVIDIA driver 580 or later. For LCC Studio the published floor is NVIDIA RTX 3060 12 GB or NVIDIA Quadro RTX A4000 16 GB, running NVIDIA driver 581.90 or later. Older 10-series cards are to be avoided regardless of memory size. Project data must be processed from a local internal disk: external drives, USB flash drives, and network paths are a documented cause of post-processing crashes.
Laptops. XGRIDS publishes no laptop-specific configuration, and the published guidance is to use a desktop for large datasets because laptops thermal-throttle under sustained load and can crash mid-reconstruction. A laptop that meets the LCC Studio basic GPU and memory figures above will process short scans. For production work on large projects, multi-segment fusion, or aerial-ground fusion, specify a desktop.
Check current XGRIDS product documentation at xgrids.com before purchasing hardware, as specifications change between product revisions.
©2026 Alpine Reality Capture LLC • XGRIDS Pro Guide™ • Site Disclaimer