M5 Max vs $2500 desktop upgrade: real-world local LLM speed comparison
vitamins1000 · reddit · 2026-09-21
A Reddit user weighs a $2500 M5 Max upgrade for local LLM use. Their M5 MacBook Air 32GB runs hot and only handles small models/low quants; a 5070 Ti desktop with 128GB DDR5 does Qwen3.8 BF16 with MTP at 20 tok/s, and more VRAM/RAM buys capacity, not speed. M5 Max's 641GB/s bandwidth wouldn't beat the desktop, just add portability — and an M7 may land in early 2027. A practical local-deployment hardware comparison thread.
More from Infra
- Turn any local LLM into a confidence-scored classifier via logprobs, full llama.cpp recipe included — DivideHorror3217 · 2026-09-21
- Redditor spins up a 4x32GB V100 vLLM server, says local setup covers 90% of work — TrailFeatures · 2026-09-21
- $5, 10-minute SFT on Qwen3.6-35B-A3B lifts GPQA +8% and MMLU-Pro +12% — josh_wills · 2026-09-21
- Agent runtime promises billions of agents and 10-20x sandbox density — astralmatrix · 2026-09-21
- SGLang team helps user debug hicache crash, earning community praise — TheZachMueller · 2026-09-21
- Two years after 'intelligence too cheap to meter', $10/$50 models are the new norm — teortaxesTex · 2026-09-21