Multiple precision modes are needed for a floating-point processing element (PE) because they provide flexibility in handling different types of numerical data with varying levels of precision and performance metrics. Performing high-precision floating-point operations has the benefits of producing highly precise and accurate results while allowing for a greater range of numerical representation. Conversely, low-precision operations offer faster computation speeds and lower power consumption. In this paper, we propose a configurable multi-precision processing element (PE) which supports Half Precision, Single Precision, Double Precision, BrainFloat-16 (BF-16) and TensorFloat-32 (TF-32). The design is realized using GPDK 45 nm technology and operated at 281.9 MHz clock frequency. The design was also implemented on Xilinx ZCU104 FPGA evaluation board. Compared with previous state-of-the-art (SOTA) multi-precision PEs, the proposed design supports two more floating point data formats namely BF-16 and TF-32. It achieves the best energy performance with 2368.91 GFLOPS/W and offers 63% improvement in operating frequency with comparable footprint and power metrics.