Grasping skill is a major ability that a wide number of real-life applications require for robotisation. State-of-the-art robotic grasping methods perform prediction of object grasp locations based on deep neural networks which require huge amount of labeled data for training and prove impracticable in robotics. In this paper, we propose to generate a large scale synthetic dataset with ground truth, which we refer to as the Jacquard grasping dataset. Specifically, the proposed Jacquard dataset builds on a subset of ShapeNet with its numerous shape objects and features at a scale of millions varied object grasping positions for a large diversity of more than 11k objects. Beyond the simulation of the grasping scene with the underlying object, the ground truth for a successful grasping position is also based on tentatives of a simulated grasping robot. We carried out experiments using an off-the-shelf CNN, with three different evaluation metrics, including real grasping robot trials. The results show that Jacquard enables much better generalization skills than a human labeled dataset thanks to its diversity of objects and grasping positions. For the purpose of reproducible research in robotics, we are releasing along with the Jacquard dataset a web interface for researchers to evaluate the successfulness of their grasping position detections using our dataset.