SDXL 1.0 VRAM Requirements
SDXL 1.0 needs about 4.1 GB of VRAM with FP8 weights at 1024×1024, loading one model at a time. In FP16 it needs 6.2 GB (10.4 GB with everything loaded), and GGUF Q4 brings it to 4.1 GB. The smallest consumer GPU that runs it is the RTX 3050 8GB with Q4 weights, or the RTX 3050 8GB in FP8.
4.1 GB peak VRAM
SDXL 1.0 at 1024×1024 with FP8 weights, loading one model at a time (ComfyUI default on smaller cards). Keeping everything loaded needs 7.3 GB.
- Diffusion model 2.4 GB
- Text encoders 0.8 GB
- VAE 0.2 GB
- Peak activations 3.1 GB
- Overhead 0.8 GB
GPU fit
- RTX 4060 8GBFitsall loaded
- RTX 3060 12GBFitsall loaded
- RTX 4060 Ti 16GBFitsall loaded
- RTX 4090 24GBFitsall loaded
- RTX 5090 32GBFitsall loaded
- RTX 6000 Ada 48GBFitsall loaded
- RTX PRO 6000 Blackwell 96GBFitsall loaded
- Apple M4 Max 128GBFitsall loaded
The VAE decode is the peak here. Turn on tiled VAE decoding to cut it to about 0.8 GB.
SDXL 1.0 VRAM by precision
At 1024×1024. Text encoders in FP8 except for the FP16 row.
| Weights | Diffusion model | Peak (swapping) | All loaded | Smallest tier |
|---|---|---|---|---|
| FP16 / BF16 | 4.8 GB | 6.2 GB | 10.4 GB | 8 GB |
| FP8 | 2.4 GB | 4.1 GB | 7.3 GB | 6 GB |
| GGUF Q8_0 | 2.5 GB | 4.1 GB | 7.4 GB | 6 GB |
| GGUF Q5_K_S | 1.7 GB | 4.1 GB | 6.5 GB | 6 GB |
| GGUF Q4_K_S / NF4 | 1.4 GB | 4.1 GB | 6.2 GB | 6 GB |
VRAM by resolution
FP8 weights, with model swapping.
| Resolution | Latent tokens | Activations | Peak VRAM |
|---|---|---|---|
| 1024×1024 | 16,384 | 0.6 GB | 4.1 GB |
| 1536×1536 | 36,864 | 1.4 GB | 8.0 GB |
About SDXL 1.0
The standard for 1024×1024 generation with a huge LoRA and checkpoint ecosystem (Pony, Illustrious, and others). The two text encoders add about 1.5 GB in FP16. Use the fp16-fix VAE: the original produces black images in FP16.
| Component | Size |
|---|---|
| Diffusion model | 2.57B parameters (UNet) |
| Text encoders | CLIP ViT-L (0.123B), OpenCLIP ViT-bigG (0.695B) |
| VAE | 0.084B parameters |
| Classifier-free guidance | Yes (2 passes per step) |
| Released | 2023-07 |
| License | CreativeML Open RAIL++-M |
SDXL 1.0 VRAM: common questions
Is 8GB enough for SDXL?
Yes. In FP16 at 1024×1024 SDXL needs about 10 GB with everything loaded, but about 6 GB when ComfyUI swaps the text encoders out during sampling, which it does automatically on 8 GB cards.
How much VRAM does SDXL 1.0 need?
At its default 1024×1024: about 6.2 GB in FP16 with model swapping (10.4 GB with everything loaded), 4.1 GB in FP8 and 4.1 GB with GGUF Q4 weights.