GPU Companies React to Stable Diffusion on a 2007

4 Просмотры
Издатель
GPU companies think you need $2,000 of hardware to run Stable Diffusion.
I used a Dell from 2007.
Python crashed before loading a single line.
So I compiled a C++ runtime by hand — and generated an image anyway.

In this video:
- Why PyTorch won't run on pre-AVX hardware — and what to do instead
- How to compile stable-diffusion.cpp with CUDA on a no-AVX CPU
- CPU inference benchmark: 389 seconds per step (yes, really)
- Fixing a broken NVIDIA driver that silently killed CUDA
- First image generated: 16.6 minutes on hardware from 2007

Hardware: Dell Dimension E520 | Core 2 Quad Q6700 | GTX 1050 Ti 4GB | 6GB DDR2
Stack: stable-diffusion.cpp | CUDA 12.2 | Ubuntu 24.04 | Stable Diffusion 1.5

Part 2 → VAE on GPU. Same machine. Under 2 minutes.

#stablediffusion #localai #linux #budgetgpu #aiart #oldpc #gpu #stabledfifusiontutorial
Категория
AI Картинки
Выполните вход или зарегистрируйтесь для отправки комментария.
Комментариев нет.