The NTIRE 2024 RAW Image Super-Resolution Challenge aimed to upscale RAW Bayer images by 2x, addressing unknown degradations such as noise and blur. A total of 230 participants registered, with 45 submitting results. The challenge dataset, based on BSRAW, included images from multiple DSLR cameras, with preprocessing steps to normalize and convert RAW images into RGGB Bayer patterns. Participants used various methods, including dual-stage networks, transformer-based models, and enhanced CNNs, to improve RAW image quality and resolution. The top-performing solutions included a dual-stage network with focal pixel loss, a dual-branch network based on HAT, and a transformer-based model (RBSFormer). These methods demonstrated significant improvements in image restoration, reducing blurriness and noise while maintaining color accuracy. The challenge also highlighted the importance of synthetic data generation and realistic degradation pipelines in training effective algorithms for RAW image super-resolution. Key findings include the effectiveness of transformer-based models in capturing long-range pixel interactions and the importance of task-specific training strategies for denoising, deblurring, and super-resolution. The results showed that RAW image super-resolution can be approached similarly to denoising, but realistic downsampling remains a challenge. The challenge emphasized the need for robust algorithms that can handle hardware-specific degradations and improve image quality in low-light and low-resolution scenarios. Overall, the challenge provided insights into the current state-of-the-art in RAW image super-resolution and the potential of deep learning methods in this domain.The NTIRE 2024 RAW Image Super-Resolution Challenge aimed to upscale RAW Bayer images by 2x, addressing unknown degradations such as noise and blur. A total of 230 participants registered, with 45 submitting results. The challenge dataset, based on BSRAW, included images from multiple DSLR cameras, with preprocessing steps to normalize and convert RAW images into RGGB Bayer patterns. Participants used various methods, including dual-stage networks, transformer-based models, and enhanced CNNs, to improve RAW image quality and resolution. The top-performing solutions included a dual-stage network with focal pixel loss, a dual-branch network based on HAT, and a transformer-based model (RBSFormer). These methods demonstrated significant improvements in image restoration, reducing blurriness and noise while maintaining color accuracy. The challenge also highlighted the importance of synthetic data generation and realistic degradation pipelines in training effective algorithms for RAW image super-resolution. Key findings include the effectiveness of transformer-based models in capturing long-range pixel interactions and the importance of task-specific training strategies for denoising, deblurring, and super-resolution. The results showed that RAW image super-resolution can be approached similarly to denoising, but realistic downsampling remains a challenge. The challenge emphasized the need for robust algorithms that can handle hardware-specific degradations and improve image quality in low-light and low-resolution scenarios. Overall, the challenge provided insights into the current state-of-the-art in RAW image super-resolution and the potential of deep learning methods in this domain.