Sidebar attacks on graphics processing units (GPUs) are a potential security threat to modern GPU architectures. We examine the feasibility of such attacks and the potential defences that could be implemented. Specifically, sidebar attacks occur through the architectural side channel of GPUs such as Open Computing Library (OpenCL), Web Graphic Library (WebGL) and Compute Unified Device Architecture (CUDA). Furthermore, we analyse GPU DOS attacks and memory exploitation attack strategies as entry into the GPU memory. In addition, the potential for operating system-level protections, such as Timeout Detection and Recovery (TDR), to detect and respond to such attacks is also analysed. Finally, a series of experiments demonstrate the efficacy of the proposed defences and the potential of sidebar attacks on GPUs. In these experiments, we found that memory isolation, Address Space Layout Randomisation (ASLR) and TDR on the operating system's timer can effectively prevent sidebar attacks on GPUs.