In the recent past, the Macroscopic Fundamental Diagram (MFD) proved to be an attractive alternative to describe the traffic states at the network level. But although complex formulations were proposed for MFD-based simulation in the literature, very few detailed validations were carried on real networks. Hence, this work focuses on a thorough validation of MFD simulators on the real field. In particular, we aim at (i) investigating the accuracy of the MFD multi-reservoir trip-based and accumulation-based models for a real large scale network by comparing the simulation results with real traffic data; and (ii) going a step further in defining proper calibration methods for the key parameters of MFD models. The network studied is the city of Lyon, which has the second greatest urban area of France, with more than 2 million inhabitants. This area is manually split into 5 reservoirs exhibiting a well defined MFD. The demand scenario was estimated for a typical weekday in a previous study based on household trip surveys and socio-demographic data. The traffic data consists of GPS trajectories of taxi fleets in Lyon, and all the loop detectors available in the area. The taxi trip data allows to determine the mean speed of each reservoir. The loop data provides a measure of the mean flow of each equipped link, which is then scaled up to each reservoir level by assuming that the measured mean flow is also representative of the non-observed links (homogeneity assumption). The average distances traveled are estimated by using shortest path calculations on the empty network (a method proposed by some of the authors in another study). A first comparison between the accumulation-based MFD simulation and the real data is presented with the evolution of accumulation in each reservoir. While providing a reliable estimation of the accumulation level in reservoir 1, the MFD simulation under-estimates the number of circulating vehicles in the other reservoirs. Different sources of errors are identified, but more investigations are still required for a better calibration of the MFD, the trip lengths and the distribution of flows.