Optimizing 3D models in the cloud with C#

By Aitor Nain Mendoza Vallejo ·August 23, 2026

Processing 3D meshes with C# in cloud environments presents a unique challenge: the files (glTF, OBJ, FBX) can weigh gigabytes, but cloud resources (like Azure Functions or Kubernetes containers) have strict memory limits. If you try to load a 2GB model directly into the RAM of a basic worker, your system will crash due to OutOfMemoryException.

In this article, we will explore key strategies for achieving cloud-based 3D model optimization without breaking the bank.

1. Streaming and MemoryMappedFiles

As we saw in our article on 3D model validation in C#, loading entire meshes into RAM generates critical pressure on the .NET Large Object Heap (LOH).

The cloud solution involves streaming. We download the blob from S3 or Azure Blob Storage directly to a temporary file on disk, and we use MemoryMappedFile to parse it lazily.

// Fragmento simplificado de streaming en Azure
var blobClient = new BlobClient(connectionString, containerName, blobName);
using var fileStream = File.OpenWrite(tempPath);
await blobClient.DownloadToAsync(fileStream);

// Procesamiento con memoria mapeada
using var mmf = MemoryMappedFile.CreateFromFile(tempPath, FileMode.Open);
// ... lógica de optimización (decimation, compresión) ...

2. Mesh Decimation

Geometric optimization is fundamental. It involves reducing the polygon count of a mesh without losing (visually) the original topology. In .NET, we can invoke native C++ libraries (using P/Invoke) specialized in geometric decimation, or use a Clean Architecture approach to isolate this logic in separate worker containers that process the edge collapse algorithm (Edge Collapse).

3. Elastic Architecture

3D model processing is CPU-bound. To avoid bottlenecks:

1. Using Azure Service Bus or RabbitMQ to queue optimization jobs.
2. Deploy workers in KEDA (Kubernetes Event-driven Autoscaling) that scale from 0 to 100 replicas based on queue size, and scale back to 0 when the job is done, saving thousands of euros per month.

Conclusion

Cloud success isn't just about writing code; it's about designing the right infrastructure. If you can isolate responsibilities with Clean Architecture and use memory intelligently, your .NET backend can process millions of 3D models per month at minimal cost.