How much VRAM does Stable Diffusion 3.5 Large need?

At BF16 with everything resident, Stable Diffusion 3.5 Large peaks at about 29.14 GB. Offloading components one at a time brings that to 16.53 GB, and quantising the weights to 4-bit brings it to 10.6 GB.

What is the cheapest GPU that runs Stable Diffusion 3.5 Large?

The GeForce GTX 1650 at around $100, running it at GGUF Q8_0 with sequential offload, taking about 2 h 21 min per generation.

Stable Diffusion · Diffusion transformer

Stable Diffusion 3.5 Large

Stable Diffusion 3.5 Large is a pipeline of 5 networks totalling 13.811 billion parameters, of which 8.146 billion do the actual generating. At 1024x1024 that is 4,096 latent tokens per denoising step.

Whole pipeline13.811B

Denoiser8.146B

Download25.7 GB

Default steps28

What the pipeline is made of

This is the part that separates a media model from a language model: these are separate networks, and whether they sit in memory together is your choice, not the model's.

Component	Parameters	Published as	On disk	What it does
transformer	8.146B	BF16	15.17 GB	Does the generating. Runs once per step, so it dominates both memory and time.
text_encoder	0.124B	BF16	0.23 GB	Turns your prompt into conditioning. Runs once, then can leave the GPU entirely.
text_encoder_2	0.695B	BF16	1.29 GB	Turns your prompt into conditioning. Runs once, then can leave the GPU entirely.
text_encoder_3	4.762B	BF16	8.87 GB	Turns your prompt into conditioning. Runs once, then can leave the GPU entirely.
vae	0.084B	BF16	0.16 GB	Converts between pixels and the compressed latent space. Small, but its decode pass is a memory spike.

These figures were derived from the published file sizes, because the repository is gated.

Memory by precision

Everything resident, at the model's native output size, on a card large enough that precision is the only constraint.

Precision	Peak VRAM	Quality	What it costs you
BF16	29.14 GB	100%	How the weights are published. No loss, and the largest footprint.
FP8	16.36 GB	97%	Halves the denoiser with a small, usually invisible cost. Needs Ada or newer.
GGUF Q8_0	17.16 GB	98%	Works on any card, unlike FP8. Slightly slower than native precision.
GGUF Q5_K_M	12.66 GB	95%	A middle step when Q8 will not fit.
GGUF Q4_K_M	11.29 GB	91%	The usual way a 12B image model gets onto an 8 GB card. Detail softens.
NF4	10.6 GB	89%	Aggressive 4-bit. Fast to load, noticeably looser on fine detail.

Memory by arrangement

The other lever: the same weights at the same precision, moved around differently. On this model quantising is the stronger lever: the denoiser is 59% of the pipeline and stays resident whatever you rearrange.

Arrangement	Peak VRAM	Time	How it works
Everything resident	29.14 GB	26 s	All components stay on the GPU. Fastest, and needs the most memory.
Text encoder on CPU	16.53 GB	27 s	The prompt is encoded once on the processor, so the encoder never touches the GPU at all. Standard practice on video models, whose encoders are often larger than their denoisers.
Component offload	16.37 GB	29 s	One component on the GPU at a time. The text encoder runs, then makes way for the denoiser. Costs a few seconds per generation.
Sequential offload	4.14 GB	2 min 32 s	Weights stream layer by layer from system RAM. Runs almost anything on almost anything, and is many times slower.

What a bigger output costs

Memory climbs with the pixel count, and the VAE decode pass climbs with it.

Output	Latent tokens	Peak VRAM	Time
512 x 512	1,024	27.25 GB	4.7 s
768 x 768	2,304	28.04 GB	12 s
1024 x 1024	4,096	29.14 GB	26 s
1536 x 1536	9,216	32.29 GB	1 min 23 s
2048 x 2048	16,384	36.69 GB	3 min 31 s

Measured at BF16 with everything resident on an RTX 4090, so the columns compare with each other rather than with your machine.

Which hardware runs Stable Diffusion 3.5 Large

118 of 118 consumer devices run it in some arrangement. Each row shows the fastest arrangement that fits on that device.

Device	Memory	Verdict	Arrangement	Peak	Time
RTX PRO 6000 Blackwell NVIDIA	96 GB	Workable	BF16 everything resident	29.14 GB	17 s
GeForce RTX 5090 NVIDIA	32 GB	Slow	BF16 everything resident	29.14 GB	20 s
RTX 6000 Ada Generation NVIDIA	48 GB	Slow	BF16 everything resident	29.14 GB	23 s
GeForce RTX 4090 NVIDIA	24 GB	Slow	GGUF Q8_0 everything resident	17.16 GB	26 s
GeForce RTX 5090 Laptop NVIDIA	24 GB	Slow	GGUF Q8_0 everything resident	17.16 GB	32 s
NVIDIA DGX Spark 128GB NVIDIA	128 GB	Slow	BF16 everything resident	29.14 GB	34 s
GeForce RTX 5080 NVIDIA	16 GB	Slow	GGUF Q5_K_M everything resident	12.66 GB	37 s
GeForce RTX 4080 SUPER NVIDIA	16 GB	Slow	GGUF Q5_K_M everything resident	12.66 GB	40 s
GeForce RTX 4090 Laptop NVIDIA	16 GB	Slow	GGUF Q5_K_M everything resident	12.66 GB	42 s
GeForce RTX 4080 NVIDIA	16 GB	Slow	GGUF Q5_K_M everything resident	12.66 GB	43 s
GeForce RTX 5080 Laptop NVIDIA	16 GB	Slow	GGUF Q5_K_M everything resident	12.66 GB	44 s
GeForce RTX 5070 Ti NVIDIA	16 GB	Slow	GGUF Q5_K_M everything resident	12.66 GB	47 s
GeForce RTX 4070 Ti SUPER NVIDIA	16 GB	Slow	GGUF Q5_K_M everything resident	12.66 GB	48 s
GeForce RTX 4070 Ti NVIDIA	12 GB	Slow	NF4 everything resident	10.6 GB	52 s
GeForce RTX 3090 Ti NVIDIA	24 GB	Slow	GGUF Q8_0 everything resident	17.16 GB	52 s
RTX A6000 NVIDIA	48 GB	Slow	BF16 everything resident	29.14 GB	54 s
GeForce RTX 4080 Laptop NVIDIA	12 GB	Slow	NF4 everything resident	10.6 GB	57 s
GeForce RTX 3090 NVIDIA	24 GB	Slow	GGUF Q8_0 everything resident	17.16 GB	59 s
GeForce RTX 4070 SUPER NVIDIA	12 GB	Slow	NF4 everything resident	10.6 GB	59 s
GeForce RTX 3080 Ti NVIDIA	12 GB	Slow	NF4 everything resident	10.6 GB	1 min 2 s
GeForce RTX 5070 NVIDIA	12 GB	Slow	NF4 everything resident	10.6 GB	1 min 8 s
GeForce RTX 3080 12GB NVIDIA	12 GB	Slow	NF4 everything resident	10.6 GB	1 min 10 s
GeForce RTX 4070 NVIDIA	12 GB	Slow	NF4 everything resident	10.6 GB	1 min 12 s
GeForce RTX 3080 10GB NVIDIA	10 GB	Slow	FP8 text encoder on cpu	8.94 GB	1 min 14 s
RTX A5000 NVIDIA	24 GB	Slow	GGUF Q8_0 everything resident	17.16 GB	1 min 15 s
GeForce RTX 2080 Ti NVIDIA	11 GB	Slow	GGUF Q8_0 text encoder on cpu	9.41 GB	1 min 21 s
GeForce RTX 5060 Ti 16GB NVIDIA	16 GB	Slow	GGUF Q5_K_M everything resident	12.66 GB	1 min 27 s
GeForce RTX 5060 Ti 8GB NVIDIA	8 GB	Barely usable	GGUF Q5_K_M text encoder on cpu	6.75 GB	1 min 31 s
GeForce RTX 4070 Laptop NVIDIA	8 GB	Barely usable	GGUF Q5_K_M text encoder on cpu	6.75 GB	1 min 34 s
GeForce RTX 4060 Ti 16GB NVIDIA	16 GB	Barely usable	GGUF Q5_K_M everything resident	12.66 GB	1 min 35 s
Jetson AGX Orin 32GB NVIDIA	32 GB	Barely usable	GGUF Q8_0 everything resident	17.16 GB	1 min 38 s
Jetson AGX Orin 64GB NVIDIA	64 GB	Barely usable	BF16 everything resident	29.14 GB	1 min 38 s
Radeon RX 7900 XTX AMD	24 GB	Barely usable	GGUF Q8_0 everything resident	17.16 GB	1 min 39 s
GeForce RTX 4060 Ti 8GB NVIDIA	8 GB	Barely usable	GGUF Q5_K_M text encoder on cpu	6.75 GB	1 min 40 s
Radeon PRO W7900 AMD	48 GB	Barely usable	BF16 everything resident	29.14 GB	1 min 40 s
GeForce RTX 3070 Ti NVIDIA	8 GB	Barely usable	GGUF Q5_K_M text encoder on cpu	6.75 GB	1 min 41 s
GeForce RTX 3070 NVIDIA	8 GB	Barely usable	GGUF Q5_K_M text encoder on cpu	6.75 GB	1 min 48 s
RTX A4000 NVIDIA	16 GB	Barely usable	GGUF Q5_K_M everything resident	12.66 GB	1 min 49 s
GeForce RTX 5060 NVIDIA	8 GB	Barely usable	GGUF Q5_K_M text encoder on cpu	6.75 GB	1 min 54 s
Radeon RX 7900 XT AMD	20 GB	Barely usable	GGUF Q8_0 everything resident	17.16 GB	1 min 58 s
Radeon RX 9070 XT AMD	16 GB	Barely usable	GGUF Q5_K_M everything resident	12.66 GB	2 min 6 s
Radeon RX 7900 GRE AMD	16 GB	Barely usable	GGUF Q5_K_M everything resident	12.66 GB	2 min 12 s
GeForce RTX 4060 Laptop NVIDIA	8 GB	Barely usable	GGUF Q5_K_M text encoder on cpu	6.75 GB	2 min 13 s
GeForce RTX 3060 Ti NVIDIA	8 GB	Barely usable	GGUF Q5_K_M text encoder on cpu	6.75 GB	2 min 15 s
GeForce RTX 4060 NVIDIA	8 GB	Barely usable	GGUF Q5_K_M text encoder on cpu	6.75 GB	2 min 24 s

Source: stabilityai/stable-diffusion-3.5-large . Downloaded 0.1 million times in the last month. See how we calculate, or browse every image model.