Did a bit of testing and thought I'd post the result here.

After a few runs on B200, I ended up with decent performance, but the setup only really made sense once I adjusted around tokens/sec.

Nothing scientific here, just practical notes from a real setup. At this point the real bottleneck is my patience.

Thought I'd post a small recipe because a lot of threads ask similar things.

This isn't revolutionary, but the combination of MLX and M3 Max felt surprisingly usable once I tuned around HBM constraints.

If someone wants the exact settings I can clean them up and post them.

This is mostly for local work rather than public serving.

This keeps coming up in conversation so I figured I'd start a thread.

It feels like the conversation has shifted hard toward inference and serving details lately, and honestly that makes sense once you start dealing with token cost.

Curious whether that lines up with what other people are seeing.

This is mostly for local work rather than public serving.

Spent some time on this today and wanted to compare notes.

I've gone back and forth between llama.cpp and LM Studio for Mistral. The interesting difference for me has been NVFP4.

Trying to avoid rebuilding the whole thing twice.

I care more about stability than squeezing the last possible tokens/sec out of it.

Quick reminder: please keep off-topic conversations in the Off Topic forum.

That helps the technical sections stay useful for people searching for hardware, models, engines, and troubleshooting threads later.

This keeps coming up in conversation so I figured I'd start a thread.

One thing I keep noticing is that people underestimate tokens/sec until they actually try to run this stuff for more than one afternoon.

Curious whether that lines up with what other people are seeing.

Has anyone else run into this?

I'm mostly working with Kimi in llama.cpp on A6000. The part I'm trying to understand is INT4.

Would appreciate practical advice rather than theory.

My use case is pretty ordinary: local chat, some document work, and occasional batch jobs.

Spent some time on this today and wanted to compare notes.

I've gone back and forth between Ollama and Ollama for Llama. The interesting difference for me has been MoE serving.

What would you pick if you had to live with it for a few months?

I care more about stability than squeezing the last possible tokens/sec out of it.

Thought I'd post a small recipe because a lot of threads ask similar things.

Maybe useful as a starting point: TensorRT-LLM on Threadripper box, keeping long context in mind from the start.

Maybe useful as a starting point for newer folks.

I care more about stability than squeezing the last possible tokens/sec out of it.

Thought I'd post a small recipe because a lot of threads ask similar things.

Maybe useful as a starting point: SGLang on mini PC, keeping continuous batching in mind from the start.

Maybe useful as a starting point for newer folks.

My use case is pretty ordinary: local chat, some document work, and occasional batch jobs.