Diabetic retinopathy (DR) is a leading cause of preventable blindness that is regularly screened for in many countries. Automated assessment of these images would substantially reduce the labour required for such screening efforts. Traditional colour fundus imaging only captures \(45^{\circ }\) of the retina, whereas modern ultra-widefield fundus images can capture up to \(200^{\circ }\) which could enable earlier detection. However, due to its novelty, there is much less research on automated methods for processing ultra-widefield images. In this manuscript, we present an ultra-fast method for automatically detecting DR and diabetic macular edema (DME) that we developed for the MICCAI Ultra-Widefield Fundus Imaging for Diabetic Retinopathy (UWF4DR) Challenge 2024 tasks 2 and 3. We use a mobile-ready, lightweight MobileNetV3 that is the fastest submission of all top 10 teams. Unconventionally, we use a single model with a single DR-risk score for both tasks, i.e. for DR and DME. We find that this DR risk score is higher in DR with DME, enabling a single predictor. Our light-weight, unified solution achieves an area under the ROC curve (AUROC) of 0.9837 for DR and 0.9785 for DME. On the official ranking, our model achieves a second place for task 3, considering speed and AUROC. We hope that our work paves the way for efficient assessment of ultra-widefield images for DR and DME. Furthermore, our unified risk score might provide an avenue towards more objective, continuous DR severity scoring. Our code is available on GitHub .