Comment by bigyabai
4 months ago
We have to get real, here - most people are not replacing GPT or Claude with local inference, even on M5. If you can afford to do that (RAM shortage or not), then you are in the minority of customers.
Alleviating the memory constraint would only really make Nvidia a danger to cloud margins, and their consumer sales are neutered while they focus on the datacenter segment. It's feels facetious to insinuate that people would be doing inference on their Macbook Neo or Wintel laptop if they only had a gorbillion gigabytes of memory and a 400W accelerator card plugged into the wall outlet.
You’re out of the loop if you don’t think m series chips with unified memory aren’t one of the best platforms for running local inference
They aren't. Apple Silicon is unusable for interactive prefill and decode speeds in agentic workflows and SOTA LLMs.
You’re just out of the loop, and that’s fine but it’s worth learning about.
There is a pretty large and growing community of us using entirely local models for our agentic flows. From GLM 4.7 flash on 32gb machines with >60tok/s to Gemma and Qwen dense and MOE models on 64gb machines all the way up to Deepseek V4 flash on 128gb machines with 450tok/s prefill and 25-30tok/s decode.
I use DS4 on the daily - it’s become my main model.
I know it’s in fashion to talk trash about Apple but their hardware outperforms other options like DGX Sparc when it comes to local inference, they got the unified memory, memory bandwidth and the GPU cores to actually be useful in a way that most other hardware just isn’t.
7 replies →