I’ve been using ChatGPT, Gemini and Claude for a long time. My work is being a Salesforce developer/admin/holyshiteverything. I’ve got an Unraid machine with an Intel i9-12900K, 64 GB of RAM, an unholy amount of storage that serves a lot of dockers like Plex. I ended up with a 7900 XT with 20 GB VRAM from a failed VM pass through experiment with a Linux project. Then I got into Claude Code wanting to make a daily RSS feed digest and then a fact checking JarvisGPT…. long story short and a 1500W APC purchase later, I’m feeling the ceiling of 20GB VRAM (also wtf qwen3 30b-a3b being 20.2 GB after KV cache fucking jerks).
I’m trying to figure out what the move is to go bigger. My mobo can’t do another full fledged GPU. But I DO have a M3 Max 36GB MacBook Pro that is my daily driver/consulting machine. Maybe the move is to sell it and try to get a 128GB one? Or maybe throw more on it and try to make it a M5 Max?
It seems from my research on here that 70B model is the size you want to be able to run. With my consulting work, it tends to deal with sensitive data. I don’t think it’s very marketable or even a good idea to send anything touching it through any cloud AI service (and I don’t). But I’d like to be able to say that I’m 100% local with all of my AI work from a privacy standpoint. But I also can’t host a data center at home and I dunno that I can run my JarvisGPT and having a coding agent at the same time on my Unraid build.
Would a good move be to try to sell my 36GB M3 Max get a M3 Max 128GB MacBook Pro as my daily driver and use it specifically for programming to have a fast response 70B coding agent?
Leave my more explorative AI work for the Unraid machine. Or does the 128GB Mac still have a lot of ceiling that are similar to what I’m hitting now? Right now, I have qwen3.5 9B as my chatbot and qwen3 30b-a3b as my overnight batch ingester as I add to my knowledge base.
[link] [comments]