We observed a sample of 90 red giant branch (RGB) stars in NGC 2808 using FLAMES/GIRAFFE and the high resolution grating with the set up HR21. These stars have previous accurate atmospheric parameters and abundances of light elements. We derived aluminium abundances for them from the strong doublet AlI 8772-8773 Angstrom as in previous works of our group. In addition, we were able to estimate the relative CN abundances for 89 of the stars from the strength of a large number of CN features. When adding self consistent abundances from previous UVES spectra analysed by our team, we gathered [Al/Fe] ratios for a total of 108 RGB stars in NGC 2808. The full dataset of proton-capture elements is used to explore in details the five spectroscopically detected discrete components in this globular cluster. We found that different classes of polluters are required to reproduce the (anti)-correlations among all proton-capture elements in the populations P2, I1, and I2 with intermediate composition. This is in agreement with the detection of lithium in lower RGB second generation stars, requiring at least two kind of polluters. To have chemically homogeneous populations the best subdivision of our sample is into six components, as derived from statistical cluster analysis. By comparing different diagrams [element/Fe] vs [element/Fe] we show for the first time that a simple dilution model is not able to reproduce all the sub-populations in this cluster. Polluters of different masses are required. NGC 2808 is confirmed to be a tough challenge to any scenario for globular cluster formation.