How should an MCX project budget memory for a display and an inference model together?

Count persistent objects and peak temporary allocations separately. Include framebuffers, model weights, tensor workspace, communication buffers, task stacks and library state. Check the actual part's memory regions and which peripherals or accelerators can access each region.

A 320 × 240 image at 16 bits per pixel needs 153,600 bytes; two such buffers need 307,200 bytes before graphics metadata. Model file size is not the same as runtime inference memory. Use tool reports and a linker map to establish both budgets.

Look for safe lifetime-based reuse, smaller display update regions or a supported lower-memory model. Track stack and heap high-water behavior under the combined workload. Do not overlap buffers merely because two separate demonstrations never happened to access them at the same time.