SDXL 1.0 is the de-facto open image-generation model — its 3.5 B parameter U-Net plus dual text encoders produces 1024×1024 images with strong composition. The ecosystem (ControlNet, LoRAs, IP-Adapter, refiner) is unmatched among open models. Runs on any 8 GB+ GPU; an RTX 4090 hits ~28 img/min at 30 steps. FLUX.1 dev has surpassed it on quality but the SDXL tooling lead remains significant.
SDXL 1.0 needs at minimum 2.1 GB of VRAM at Q4_K_M quantization (7.7 GB at FP16). The cheapest GPU that comfortably fits with KV-cache headroom is the Intel Arc A380 6GB (6 GB VRAM, $139 MSRP). Measured throughput hits 52 img/min on NVIDIA B200 192GB. This model fits a single consumer GPU under 48 GB, so a one-card build works.
Weights only. Add ~20-30% for KV cache at typical context lengths.
| Quant | VRAM | Quality | Notes |
|---|---|---|---|
| FP16 | 7.7 GB | Reference | Training-precision weights |
| Q8_0 | 3.9 GB | Near-lossless | 8-bit, ~0.1% perplexity hit |
| Q6_K | 3.1 GB | Very high | 6-bit, near Q8 quality |
| Q5_K_M | 2.7 GB | High | Strong middle ground |
| Q4_K_M | 2.1 GB | Balanced (recommended) | Default for local deployments |
| Q4_0 | 2.2 GB | Legacy 4-bit | Older GGUF, kept for compatibility |
| Q3_K_M | 1.7 GB | Lossy | When VRAM is very tight |
| Q2_K | 1.2 GB | Extreme | Rescue option, quality degrades visibly |
Filtered from MyAIHardware's GPU database. FP16 needs 7.7 GB; Q4_K_M needs 2.1 GB.
30 measurement(s) for this model from MyAIHardware's benchmark database.
| Device | Speed | Quant | Context | Power |
|---|---|---|---|---|
| NVIDIA B200 192GB | 52 img/min | FP16 | 0 | 1000W |
| NVIDIA GeForce RTX 5090 32GB | 38 img/min | FP16 | 0 | 575W |
| NVIDIA H200 141GB | 36 img/min | FP16 | 0 | 700W |
| NVIDIA GeForce RTX 5090 32GB | 32 img/min | FP16 | 0 | 575W |
| NVIDIA H100 SXM5 80GB | 28 img/min | FP16 | 0 | 700W |
| NVIDIA GeForce RTX 4090 24GB | 24 img/min | FP16 | 0 | 450W |
| AMD Instinct MI300X 192GB | 22 img/min | FP16 | 0 | 750W |
| NVIDIA RTX 6000 Ada 48GB | 22 img/min | FP16 | 0 | 300W |
| AMD Instinct MI300X 192GB | 22 img/min | FP16 | 0 | 750W |
| NVIDIA L40S 48GB | 18 img/min | FP16 | 0 | 350W |
| NVIDIA GeForce RTX 5080 16GB | 18 img/min | FP16 | 0 | 360W |
| NVIDIA GeForce RTX 5090 32GB | 18 img/min | FP16 | 0 | 575W |
| NVIDIA GeForce RTX 4080 Super 16GB | 16 img/min | FP16 | 0 | 320W |
| NVIDIA GeForce RTX 5070 Ti 16GB | 14 img/min | FP16 | 0 | 300W |
| AMD Radeon RX 7900 XTX 24GB | 13 img/min | FP16 | 0 | 355W |
| NVIDIA GeForce RTX 3090 24GB | 12 img/min | FP16 | 0 | 350W |
| NVIDIA GeForce RTX 4070 Ti 12GB | 11 img/min | FP16 | 0 | 285W |
| NVIDIA GeForce RTX 5070 12GB | 10.5 img/min | FP16 | 0 | 250W |
| NVIDIA GeForce RTX 4070 12GB | 9 img/min | FP16 | 0 | 200W |
| NVIDIA GeForce RTX 3090 24GB | 7.5 img/min | FP16 | 0 | 350W |
Official weights + popular runtime tags.
Step-by-step guides that use this model.
Models in a similar size or capability class.
Use these in papers, blog posts, or internal docs.
@misc{sdxl-1-0-2023,
title={SDXL 1.0},
author={Stability AI},
year={2023},
url={https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0}
}Stability AI (2023). SDXL 1.0 [Model card]. Hugging Face. https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0
SDXL 1.0 (Stable Diffusion, Stability AI, 2023). Available at https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0.