SDXL on a Core i5-650 from 2010, 8 GB DDR3 RAM and an RX 570 with 4 GB VRAM from 2017 :O Model: sdxlMergeheaven_betaM15.safetensors 4-bit quantization performed directly by stable-diffusion.cpp Backend: Vulkan Resolution: 512 × 512 Generation time: 574.92 seconds Afterwards, the image was upscaled with Real-ESRGAN, also using Vulkan. Old hardware, especially older AMD GPUs, is proving to be surprisingly capable for local AI. This PC was built from clearance parts in early 2012 for a little over €300. The ASUS P7P55D-E motherboard dates back to 2009 and the Core i5-650 is from 2010. Even the 8 GB of RAM felt like an absurd amount at the time. The RX 570 was a €40 used GPU I bought last year xD And now, in 2026, this machine is running SDXL locally. Who knows what else it can do. #EdgeAI #EdgeComputing #AI #LocalAI #AIOnTheEdge #Upcycling #HardwareUpcycling #SustainableComputing #EfficientComputing #FOSS #OpenSource #Linux #Vulkan #AMD #RX570 #SDXL #StableDiffusion #RealESRGAN #OldHardwa...
SDXL on a Core i5-650 from 2010, 8 GB DDR3 RAM and an RX 570 with 4 GB VRAM from 2017 :O Model: sdxlMergeheaven_betaM15....
