Simulation · FEA Lab

Structural simulation. On your Mac’s GPU.

FEA Lab provides finite element analysis on Apple Silicon Macs: linear static, nonlinear static, modal and linear buckling studies on regular HEXA8 grids. Metal GPU and supported CPU routes produce stress, displacement and solver records, with limitations stated alongside results.

Start 7-day trial

Read the limits

All four shipped in Lattiform

FEA Lab in Lattiform showing von Mises stress on an 80 by 20 by 20 millimetre aluminium cantilever, with a stress legend from 0.3 to 5.7 MPa — Actual app capture

Four analysis types, each with its limits.

Linear static

Deflection, stress and strain under steady loads, with named load cases. Solved with Metal in single precision with the final residual checked in double, or on the CPU in double precision.

Limits: Assumes small deflections and a linear elastic material.

Evidence: reports V3 and V4 tested the former FEA Preview’s linear static route in September 2026, not this build.

See verification evidence ↗

Nonlinear static

Large displacement: equilibrium on the deformed shape, reached in load steps solved by Newton’s method. St Venant–Kirchhoff elastic material, on the CPU in double precision or with its main work on the GPU.

Limits: Elastic only in FEA Lab. Loads keep their original direction, supports hold zero displacement, and a step that will not converge stops the solve.

Evidence: report V6 tested a nonlinear solver source revision, not a shipped build.

See verification evidence ↗

Linear buckling

Critical load factors and buckling mode shapes for a linear elastic part under the study’s loads, solved on the GPU. Mode shapes are drawn at an arbitrary scale.

Limits: Whole grid cells only. Supports hold zero displacement and loads keep their direction and grow in proportion. No post-buckling, follower loads or imperfections.

No published verification report for this capability in the current build. Internal development checks do not verify the shipped package.

See evidence status ↗

Every result explains itself

The solver’s own record, beside the picture.

Each result shows how it was reached: convergence, the relative residual against the tolerance you asked for, force and moment balance, the backend that solved it and the solve time. Warnings are shown, not hidden.

FEA Lab in Lattiform with the Solution quality panel open, showing convergence, relative residual, force and moment balance, backend and solve time for the cantilever study — Actual app capture

Local solving and the analysis grid.

Shipped

Solved on your Mac

Studies run on the GPU through Metal or on the CPU. Models, geometry and results are not sent anywhere to be solved. A linear static study that would need more than half of your Mac’s memory is refused before it starts.

Shipped

A regular grid of cube elements

Every study runs on a regular, axis-aligned grid of 8-node cube elements with partly filled cells at the surface. You choose the element size; curved and inclined surfaces are approximated by the grid.

Shipped

Refinement and convergence

Refine an interior region 2× or 4× in linear static studies, and run a 1×/2×/4× mesh convergence sequence at a probe point to see whether a value has settled.

FEA Lab in Lattiform showing a STEP part on its analysis grid of cube elements — Actual app capture

Beta

Feature-aware element sizing

Element sizes chosen from the part’s own fillets, holes, thin walls and loaded faces, offered as advice in FEA Lab and used by Local refinement. In the current version; not yet verified.

Beta

Adaptive mesh refinement

Refinement where the estimated error is largest, on the CPU and Metal, for linear static studies. In the current version, off unless you turn it on, and not yet verified.

What FEA Lab does not do.

Physics not offered

No transient, shock, random-vibration or response-spectrum analysis, and no fatigue, heat transfer, thermal stress or fluid flow. No anisotropic or composite materials.

Materials and contact

Nonlinear static uses an elastic material in FEA Lab’s settings. Plasticity and contact with a rigid surface exist only through the agent (MCP) tools, as a beta without published evidence.

Model

One material per study. Several bodies are welded into one, with no bonded interfaces or joints. Only cube elements: no tetrahedral, shell or beam elements.

Accuracy at surfaces

Stress read at curved and inclined surfaces is an approximation; refine and use the convergence sequence. Stress at a sharp inside corner or a point load does not settle as the grid is refined.

Results do not certify a design for production, safety or regulatory use. Full scope and known faults for the current version. Verification evidence.

Status of every simulation capability.

Linear static

Shipped

In Lattiform now

Deflection, stress and strain under steady loads, with named load cases. Solved on the GPU with Metal (single precision, final residual checked in double) or on the CPU in double precision.

Small deflections and a linear elastic material.

Nonlinear static (large displacement)

Shipped

In Lattiform now

Equilibrium on the deformed shape, reached in Newton-solved load steps. St Venant–Kirchhoff elastic material. CPU in double precision, or with its main work on the GPU.

Elastic material only in FEA Lab. Loads keep their original direction. Supports must hold zero displacement. A step that will not converge stops the solve.

Modal

Shipped

In Lattiform now

Undamped natural frequencies and mode shapes, 1 to 128 modes, with mass participation, a two-mesh convergence comparison and a base-excitation estimate at a damping ratio you choose. GPU only.

Refuses a part whose surface leaves grid cells partly filled, which includes most imported and Model Prep parts. Ignores loads and gravity.

Linear buckling

Shipped

In Lattiform now

Critical load factors and buckling mode shapes for a linear elastic part under the study's loads, solved on the GPU. Each result is checked again in double precision, and an inertia check confirms the lowest mode where it applies; otherwise the lowest load factor is an upper bound.

Whole grid cells only (most imported and Model Prep parts are refused). Supports must hold zero displacement; loads must keep their direction and grow in proportion. No nonlinear prestress, follower loads, post-buckling or imperfections.

Analysis grid

Shipped

In Lattiform now

A regular, axis-aligned grid of 8-node cube (HEXA8) elements with partly filled cells. You choose the element size. Interior 2× or 4× local refinement for linear static, and a 1×/2×/4× mesh convergence sequence at a probe.

Curved and inclined surfaces are approximated by the grid. No tetrahedral, shell or beam elements.

Feature-aware element sizing

Beta

In Lattiform now

Element sizes taken from the part's own fillets, holes, thin walls and loaded faces, offered as advice in FEA Lab's Feature sizing panel and used by Local refinement.

Advice only until you apply it. No published evidence.

Adaptive mesh refinement

Beta

In Lattiform now

Local refinement in linear static studies: splits cubes at the part's bores, fillets and thin walls and where the estimated error is largest, on the CPU and Metal, and merges quiet regions back.

Off unless you turn it on, and not yet verified: no published evidence.

Plasticity and rigid-surface contact

Beta

In Lattiform now

J2 plasticity, Neo-Hookean material and up to three frictionless or frictional rigid resting faces in a nonlinear study.

Only through the agent (MCP) design tools, not FEA Lab's own settings. No published evidence.

Local Metal solving

Shipped

In Lattiform now

Studies are solved on the Mac's GPU through Metal, or on the CPU. Models, geometry and results are not sent anywhere to be solved.

A linear static study whose estimated memory need exceeds half of the Mac's memory is refused before solving.

Run your first study.

Choose Student, Individual or Commercial. Each is a yearly licence with a 7-day free trial.

Start 7-day trial

See the evidence