The new VLTI (Very Large Telescope Interferometer) 1 instrument GRAVITY<SUP>5, 22, 23</SUP> is equipped with a fringe tracker<SUP>16</SUP> able to stabilize the K-band fringes on six baselines at the same time. It has been designed to achieve a performance for average seeing conditions of a residual OPD (Optical Path Difference) lower than 300 nm with objects brighter than K = 10. The control loop implementing the tracking is composed of a four stage real time computer system compromising: a sensor where the detector pixels are read in and the OPD and GD (Group Delay) are calculated; a controller receiving the computed sensor quantities and producing commands for the piezo actuators; a concentrator which combines both the OPD commands with the real time tip/tilt corrections offloading them to the piezo actuator; and finally a Kalman<SUP>15</SUP> parameter estimator. This last stage is used to monitor current measurements over a window of few seconds and estimate new values for the main Kalman<SUP>15</SUP> control loop parameters. The hardware and software implementation of this design runs asynchronously and communicates the four computers for data transfer via the Reflective Memory Network<SUP>3</SUP>. With the purpose of improving the performance of the GRAVITY<SUP>5, 23</SUP> fringe tracking<SUP>16, 22</SUP> control loop, a deviation from the standard asynchronous communication mechanism has been proposed and implemented. This new scheme operates the four independent real time computers involved in the tracking loop synchronously using the Reflective Memory Interrupts<SUP>2</SUP> as the coordination signal. This synchronous mechanism had the effect of reducing the total pure delay of the loop from 3.5 [ms] to 2.0 [ms] which then translates on a better stabilization of the fringes as the bandwidth of the system is substantially improved. This paper will explain in detail the real time architecture of the fringe tracker in both is synchronous and synchronous implementation. The achieved improvements on reducing the delay via this mechanism will be quantified.