Robotic grasping is a fundamental skill required for object manipulation in robotics. Multi-fingered robotic hands, which mimic the structure of the human hand, can potentially perform complex object manipulations. Nevertheless, current techniques for multi-fingered robotic grasping frequently predict only a single grasp for each inference time, limiting their versatility and efficiency. This paper proposes a differentiable multi-fingered grasp generation network (DMFC-GraspNet) with two main contributions to address this challenge. Firstly, a novel neural grasp planner is proposed, which predicts a new grasp representation to enable versatile and dense grasp predictions. Secondly, a scene creation and label mapping method is developed for dense labeling of multi-fingered robotic hands, which allows a dense association of ground truth grasps. The proposed approach is evaluated through simulation studies and compared to existing approaches. The results demonstrate the effectiveness of the proposed approach in predicting versatile and dense grasps, and in advancing the field of robotic grasping.