Big GPUs can perform well without the need for large PCs, as shown by tests comparing Raspberry Pi 5 setups to modern desktop systems. The Raspberry Pi connects to external GPUs via a single PCIe Gen 3 lane, significantly lower than the multiple lanes available in typical desktops. Despite this limitation, the Pi can still match or even exceed the performance of higher-end PCs in specific scenarios, particularly in efficiency and cost. The Raspberry Pi eGPU setup costs between $350-$400, while an Intel PC setup ranges from $1500-$2000. Testing focused on various applications, including media transcoding with Jellyfin, 3D rendering with GravityMark, and AI performance using multiple graphics cards. In media transcoding, the Intel PC outperformed the Pi in raw data throughput but the Pi handled typical transcoding tasks smoothly, especially with H.264 and H.265 files. The GravityMark benchmark revealed that while Intel’s desktop was generally faster, the Pi edged out the PC with an older AMD RX 460, showcasing strong efficiency in performance per watt. AI performance comparisons showed mixed results. The AMD Radeon AI Pro R9700 struggled on the Pi, but when tested with an Nvidia RTX 3060, the Pi performed comparably, sometimes even better, on medium-sized models. Notably, the Pi achieved 11.83 tokens per second for Llama 3 70b, only slightly behind a modern server with the same GPU setup. This illustrates the potential of using a Raspberry Pi for GPU tasks, especially for those needing cost-effective solutions without sacrificing much performance.