I ran local LLMs on my phone for a week, and now my desktop setup feels like overkill

Question

After a week of running local LLMs exclusively on an iPhone, the author found mobile performance surprisingly close to desktop. The Gemma E2B (5B parameters) on phone handled chat, brainstorming, and even multimodal image tasks nearly as well as the 8B desktop variant. The main limitations were smaller context windows and heavy document processing. The experiment revealed that for most everyday AI tasks, the desktop setup was overkill — the author ended up reaching for the phone more often than LM Studio or llama.cpp on desktop.