pulled deepseek-r1:32b on my M2 Pro last night "just to see". Activity Monitor hit ~18GB before the first decent answer showed up, fans sounded like a leaf blower aimed at my face.
Q4 was usable for a short refactor. Q5 started swapping and Cursor lagged behind the chat window. anyone actually running the bigger tags on 32GB without closing Chrome?
Pulled qwen2.5 coder:32b q4 K M into Ollama on an M2 Max (64GB). Activity Monitor said 41GB before the refactor even started. Halfway through rewriting a messy FastAPI router it just… died. No stack…
Spent last week on Continue + Ollama ( qwen2.5 coder:14b ) for a Rust CLI refactor. Config clearly pointed at local. Half the turns still billed Anthropic until I noticed the usage graph. Free Contin…
Spent Friday night wiring Needle 2 into a little home assistant bridge on a Pi 5. The 14 MB binary is real — session RAM sat around 30 MB, and plain tool calls (turn light on/off, read a sensor) were…
5 comments
Join the discussion
Log in to comment.
which exact tag?
deepseek-r1:32b-q4_K_Mvs a generic32bpull behaves wildly different on Apple silicon.also 18GB before first token is normal if context defaults are high. what does
ollama psshow for size + processor?yeah pull the exact tag.
ollama pull deepseek-r1:32b-q4_K_Mvs bare32bis night and day on M2.i also pin with:
in a Modelfile so Activity Monitor stops lying about "idle".
same machine here. i keep
num_ctxat 4k for day-to-day and quit Chrome before the pull. Q5 on 32GB only works if nothing else is hungry.if Cursor starts hitching, drop to Q4 and bump
num_gpulayers — fans still spin but swap stops thrashing.same. i quit Slack + Figma before any pull now. last time I left both open, Cursor's autocomplete started typing mid-sentence like it was buffering a youtube ad.
Q4 + num_ctx 4k is the only combo that lets me keep a design review tab open without the fans auditioning for a leaf blower ad.
on a 64GB linux box deepseek-r1:32b-q4_K_M is fine. on 32GB apple silicon i treat Q5 as a weekend experiment, not a daily driver.
the leaf-blower fan is usually the KV cache + metal allocator fighting Chrome tabs, not the weights alone.
ollama show deepseek-r1:32b --modelfileand check if someone left num_ctx at 32k.