Old hardware doesn't have to mean obsolete! I got both Ollama and Stable Diffusion running on my secondary PC using the Vulkan backend: Intel Core i5-650 (2010) | 8 GB DDR3 RAM | AMD Radeon RX 570 (4 GB VRAM) (2017) Results: • Ollama: 5.5 tokens/s • Stable Diffusion 1.5: 125.62 s for a 512×768 image (20 steps) I tested Google's Gemma 3 4B (4-bit) locally. It is obviously not a large model, but it already works surprisingly well as a small tutoring assistant. For example, it can explain math rules like the product rule step by step instead of just giving an answer. I built stable-diffusion.cpp from source. Once that was done, everything worked surprisingly well. Source: https://github.com/leejet/stable-diffusion.cpp It won't break any speed records, but it's a nice reminder that local AI is still accessible on aging hardware. #AI #LocalAI #Ollama #Gemma3 #StableDiffusion #OpenSource #FOSS #Linux #AMD #Vulkan #SelfHosted #MachineLearning
Old hardware doesn't have to mean obsolete! I got both Ollama and Stable Diffusion running on my secondary PC using the ...
