Synthetic aperture radar (SAR) is an active remote sensing tool that collects data on an area using large wavelength pulses. SAR produces images which can then be passed to conventional analysis algorithms. This data can be used for studying topographical changes to the earth, monitoring vege tational development, and in machine-learning (ML) tasks like object detection. However, these images contain a substantial amount of noise, known as speckle, that can inhibit the accuracy of post-processing efforts. While conventional despeckling relies on statistical techniques, there is a growing interest in ML based methods due to their state-of-the-art accuracy. Running despeckling algorithms onboard satellite payloads enables quick, real-time decisions about a sensed area. However, the runtime performance of ML models is limited by onboard compute capabilities. Additionally, the large size of SAR data makes it difficult to downlink. Instead, improving model runtime through quantization can enable real-time performance without data transfer. This research examines the impact of quantization on common SAR despeckling models. We compare the performance of quantized models with their full-precision counterparts on both synthetic and real SAR data. Quantized models achieve up to a 4.66× speedup with up to a 4× decrease in memory usage over full-precision models. PSNR accuracy loss with quantization on synthetic data can be as little as 0.062 dB. Quantization performance with real SAR data remains comparable to full precision models except in homogeneous regions, where ENL can decrease by up to 82.86%.