This paper reviews the NTIRE 2024 RAW Image Super-Resolution Challenge, highlighting the proposed solutions and results. The challenge aims to upscale RAW Bayer images by 2x while considering unknown degradations such as noise and blur. A total of 230 participants registered, and 45 submitted results. The performance of the top-5 submissions is reviewed to gauge the current state-of-the-art in RAW Image Super-Resolution. RAW Image Super-Resolution is an active research direction, focusing on upscaling hardware-specific RAW image representations while dealing with hardware characteristic properties. The lack of standardization in camera Image Processing Signal (ISP) implementation introduces variety into the camera market, with corrections made through image processing algorithms to overcome hardware limitations. RAW images represent the discretized and quantized representation of the image signal, and their processing poses significant advantages over standard sRGB representation in various applications like image denoising, deblurring, exposure adjustment, and super-resolution. The challenge dataset is based on BSRAW, using images from the Adobe MITS5K dataset, which includes images from multiple Canon and Nikon DSLR cameras. The DSLR images are manually filtered to ensure diversity and natural properties. The pre-processing involves normalizing all RAW images and converting them into the RGGB Bayer pattern. Training data consists of 1064 1024×1024×4 clean high-resolution (HR) RAW images, with low-resolution (LR) degraded images generated during training using a degradation pipeline. The challenge evaluates three testing splits: Validation (40 images), Test 1MP (200 images), and Test at full-resolution (12MP). Participants process the corresponding LR RAW images and submit their results. The paper provides detailed visual comparisons and discusses the overall results, highlighting the improvements in image quality and resolution. The paper presents the solutions submitted for the challenge, including detailed descriptions of the best solutions. It also provides information on the challenge setup, dataset, and evaluation metrics. The top-performing solutions improve the baseline performance, with neural networks being more complex in design and computation. The paper discusses the challenges and future directions in RAW image super-resolution, emphasizing the importance of realistic data synthesis and the need for more realistic downsampling techniques.This paper reviews the NTIRE 2024 RAW Image Super-Resolution Challenge, highlighting the proposed solutions and results. The challenge aims to upscale RAW Bayer images by 2x while considering unknown degradations such as noise and blur. A total of 230 participants registered, and 45 submitted results. The performance of the top-5 submissions is reviewed to gauge the current state-of-the-art in RAW Image Super-Resolution. RAW Image Super-Resolution is an active research direction, focusing on upscaling hardware-specific RAW image representations while dealing with hardware characteristic properties. The lack of standardization in camera Image Processing Signal (ISP) implementation introduces variety into the camera market, with corrections made through image processing algorithms to overcome hardware limitations. RAW images represent the discretized and quantized representation of the image signal, and their processing poses significant advantages over standard sRGB representation in various applications like image denoising, deblurring, exposure adjustment, and super-resolution. The challenge dataset is based on BSRAW, using images from the Adobe MITS5K dataset, which includes images from multiple Canon and Nikon DSLR cameras. The DSLR images are manually filtered to ensure diversity and natural properties. The pre-processing involves normalizing all RAW images and converting them into the RGGB Bayer pattern. Training data consists of 1064 1024×1024×4 clean high-resolution (HR) RAW images, with low-resolution (LR) degraded images generated during training using a degradation pipeline. The challenge evaluates three testing splits: Validation (40 images), Test 1MP (200 images), and Test at full-resolution (12MP). Participants process the corresponding LR RAW images and submit their results. The paper provides detailed visual comparisons and discusses the overall results, highlighting the improvements in image quality and resolution. The paper presents the solutions submitted for the challenge, including detailed descriptions of the best solutions. It also provides information on the challenge setup, dataset, and evaluation metrics. The top-performing solutions improve the baseline performance, with neural networks being more complex in design and computation. The paper discusses the challenges and future directions in RAW image super-resolution, emphasizing the importance of realistic data synthesis and the need for more realistic downsampling techniques.