I Ran DeepSeek R1 on a $80 Pi vs $250 Jetson vs $1000 Mac — Here’s What Happened

A hands-on comparison of running DeepSeek R1 (1.5B parameter model) locally on three devices: a Raspberry Pi 5 ($80), Nvidia Jetson Orin Nano ($250), and MacBook Air M3 ($1,000), using Ollama. The MacBook achieves 72 tokens/sec, the Jetson 22 tokens/sec, and the Pi 9 tokens/sec. All three produce equivalent output quality since they run identical model weights — only inference speed differs. The Pi is 12.5x cheaper than the Mac but 9x slower. The post also covers model quantization (Q4 vs Q8) and parameter size trade-offs for edge deployment.