I am an OG nerd started with a Commodore 64 in the early 80’s and never looked back and I have just gotten into Agents and Vibe coding, in a week I made multiple games, reprogrammed my TV so basically its Anarchy at my house…not out of the norm..
I have 3 pc’s 7900xtx, 9070xt and 9060xt and no CUDA, unfortunately that runs more models that I would like to. I was forced medically to retire years ago and I have had 20 years waiting for something like this to happen, is it worth the shaft of the NVIDIA tax to have local AI for most people probably not being honest, (for many, many reasons I want to do everything offline)
- I really need the other side of the rainbow to do everything I want in AI, is there a reasonable trade off or am I spending 10 grand?
- is there something I can do with what I have to get around the CUDA price wall?
Thanks
You already have the useful card. The 7900 XTX’s 24 GB is what fits a local coding model, and llama.cpp on Vulkan runs on that GPU without a CUDA card. Point the agents at that box and leave the 9070 XT and 9060 XT for the displays and games. A $10k NVIDIA machine is for a model that will not fit in 24 GB, or for a tool that is still CUDA-only after you try the Vulkan build. For one person doing this offline, I would spend a weekend on llama.cpp before I spent the money.