I'm interested in running AI models locally for chatting, writing code, and possibly generating videos. I have a MacBook with an M3 chip and 8GB of unified memory. I've heard that using an SSD or a 512GB flash drive might help with AI workloads, and I also have an older Dell laptop whose storage I could use. I'm not very technical, so I'm trying to understand whether I can run two different models on the MacBook, whether external storage would make a meaningful difference, and if there's a practical way to use the older laptop too.
4 Answers
The old Dell’s storage can’t directly add memory to the MacBook. You could install and run a model on the Dell separately if its processor, RAM, and graphics hardware are suitable, but combining both laptops into one local AI system is complicated and usually not worthwhile for a beginner. Start with one small, quantized chat or coding model on the MacBook and see how it performs before attempting anything more ambitious.
With 8GB of unified memory, you can experiment with small, compressed language models for basic chat or coding, but performance may be slow and you’ll need to close other applications. Running two models at the same time is unlikely to be practical because they would compete for the same memory. You could run one, stop it, and then switch to the other.
External storage can hold the model files, but it does not replace RAM or unified memory. Once a model is loaded, the computer needs enough memory for the model and its working data. An external SSD may improve loading times compared with a hard drive, but it won’t make an 8GB machine capable of running models that require much more memory.
So the SSD would mainly help with storage and loading, but not with the actual amount of memory available while the model is running?
Local video generation is much more demanding than text chat or coding. On an 8GB MacBook, most video models will either fail to run, require extreme compromises, or take an impractical amount of time. A small text model is a reasonable project, but video generation is generally better handled by a more powerful computer or a hosted service.
That makes sense. I was hoping to keep both available, but running them one at a time may be more realistic.