ESO has now delivered or tested in-house, four new 5.3 μm cut-off H2RG detectors, for various projects, such as MATISSE for the VLTI and the upgrade project for CRIRES, the cryogenic high-resolution infrared echelle spectrograph for the VLT. The specified instruments have required the implementation of some of the more unusual read out options for these detectors, which may have already been used by other groups, for example, the line-reset-read and line-read-reset modes rather than the standard global reset mode. The detectors are also offered with both output speed options, that is, the standard slow, low noise readout and the faster, higher noise readout, where > 10 frames/s are possible. In the process of building these detector systems and implementing these new options we have delved deeper into some of the lesser known features of these detectors and tried to characterize them more fully. It is important that these characteristics are well understood before delivery of the next generation of detectors for the ELTs where high speed and windowing options are required. We obtain very good performance at 2 Mpixel/s pixel speeds with less than 40 e- rms read noise, in all other aspects such as linearity, noise versus number of non-destructive reads and cross talk then the performance of the outputs is the same as slow speed operation. However, the high speed output stages are quite complex to operate, they need to be very well tuned and are prone to oscillation, if not set correctly. We will report on the best bias options to optimize their performance. Some stability issues are also seen with the slow outputs and this is also reported. Likewise we have observed differences between global reset and line reset for the detectors, manifested in a significant increase in detector full well for the line reset option, this also will be reported on. We have also determined that there may be signal induced by the detector readout clocking process for certain detector material or ROIC revision, at a significant level such that this may be the probable limiting factor of why Fowler sampling reaches a minimum noise value of approximately 3 e- rms for a small number of reads and then increases with further non-destructive reads.