Diagnostics at ASDEX Upgrade have available a very large number of highly developed measuring channels. The prospect of making this wealth of information usable for plasma optimization led to the implementation of a number of diagnostics running data acquisition in real-time (RT). Ultimately, this development aims to achieve a network of intelligent diagnostics delivering analysed data for high-level plasma performance control such as profile shaping and NTM stabilization.The new RT diagnostics consist of standard industrial 19" servers organized in clusters and running a standard UNIX multiprocessor RT-capable operating system (RT OS). Built-to-purpose computer interface cards deliver data (e.g. via serial links) from the data acquisition (DAQ) front-ends directly into the main memory of the DAQ servers. An RT data analysis task immediately following the running direct memory access (DMA) data transfers processes the data and delivers the results to follow-up systems in the control chain.Whereas the first systems were implemented in a simple JBOC (just a bunch of computers) configuration being operated as a number of single diagnostics, newer systems are integrated into diagnostic clusters using parallel computing techniques such as message passing interface (MPI).The paper describes the hardware (ADC front-ends, serial I/O, selection criteria and performance of the involved computer busses and systems) and software (DAQ, DA, RT OS, MPI) architecture of the assembled systems. Benchmark results for DAQ and MPI bandwidth and latencies as well as for the behaviour of the RT OS will be given.