Practical GPU Choices for Earth Observation: ResNet-50 Training Throughput on Integrated, Laptop, and Cloud Accelerators

TL;DR

This paper benchmarks ResNet-50 training throughput on different GPUs for land cover classification, demonstrating significant speed-ups on consumer and cloud GPUs while maintaining accuracy, thus enabling scalable geospatial analytics.

Contribution

It provides a comprehensive comparison of GPU performance for ResNet-50 training in earth observation tasks, highlighting practical choices for deployment.

Findings

Up to 2x training speed-up on RTX 3060 and Tesla T4

Maintained high classification accuracy on EuroSAT dataset

Demonstrated feasibility of using consumer and cloud GPUs for scalable geospatial analytics

Abstract

This project implements a ResNet-based pipeline for land use and land cover (LULC) classification on Sentinel-2 imagery, benchmarked across three heterogeneous GPUs. The workflow automates data acquisition, geospatial preprocessing, tiling, model training, and visualization, and is fully containerized for reproducibility. Performance evaluation reveals up to a 2x training speed-up on an NVIDIA RTX 3060 and a Tesla T4 compared to the Apple M3 Pro baseline, while maintaining high classification accuracy on the EuroSAT dataset. These results demonstrate the feasibility of deploying deep learning LULC models on consumer and free cloud GPUs for scalable geospatial analytics.