The rapid expansion of IoT systems and the corresponding surge in network traffic have increased the demand for efficient and low-latency Intrusion Detection Systems (IDS). Although neural network-based IDS models demonstrate strong performance in identifying malicious patterns, deploying these models on embedded systems (edge devices) remains challenging due to resource and power constraints. This work proposes IDSPfree, an FPGA-based low resource consumption and high energy-efficient system for intrusion detection. In particular, we first introduce 8-bit floating point (FP8) quantization to reduce data-width but maintain accuracy. Building on the FP8 data format, we further develop the optimized FP8 approximate multipliers and symmetric approximate activation functions. Experimental results show that IDSPfree maintains an average accuracy loss below 2\% on three typical datasets. Additionally, IDSPfree achieves average performance improvements of 9.0× and 3.83×, with up to 6.13× and 9.11× energy consumption saving compared to edge CPU and edge GPU, respectively. Moreover, compared to edge-FPGA-based IDS designs, IDSPfree achieves the lowest power consumption while delivering comparable performance.