How much VRAM does Stable Video Diffusion XT need?

At BF16 with everything resident, Stable Video Diffusion XT peaks at about 13.89 GB. Offloading components one at a time brings that to 13.89 GB, and quantising the weights to 4-bit brings it to 10.98 GB.

What is the cheapest GPU that runs Stable Video Diffusion XT?

The GeForce GTX 1080 Ti at around $150, running it at BF16 with sequential offload, taking about 83 h 39 min per generation.

Stable Diffusion · UNet

Stable Video Diffusion XT

Stable Video Diffusion XT is a pipeline of 3 networks totalling 2.255 billion parameters, of which 1.525 billion do the actual generating. At its native 1024x576 and 25 frames the denoiser works on 230,400 latent tokens at once, and attention over that sequence is what sets the render time.

Whole pipeline2.255B

Denoiser1.525B

Download8.4 GB

Default steps25

Native length25 frames

Frame rate6 fps

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
unet	1.525B	F32	5.68 GB	Does the generating. Runs once per step, so it dominates both memory and time.
vae	0.098B	F32	0.36 GB	Converts between pixels and the compressed latent space. Small, but its decode pass is a memory spike.
image_encoder	0.632B	F32	2.35 GB	Auxiliary encoder.

These figures were measured from the weight file headers.

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	13.89 GB	100%	How the weights are published. No loss, and the largest footprint.
FP8	11.88 GB	97%	Halves the denoiser with a small, usually invisible cost. Needs Ada or newer.
GGUF Q8_0	12.01 GB	98%	Works on any card, unlike FP8. Slightly slower than native precision.
GGUF Q5_K_M	11.3 GB	95%	A middle step when Q8 will not fit.
GGUF Q4_K_M	11.08 GB	91%	The usual way a 12B image model gets onto an 8 GB card. Detail softens.
NF4	10.98 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 68% of the pipeline and stays resident whatever you rearrange.

Arrangement	Peak VRAM	Time	How it works
Everything resident	13.89 GB	27 s	All components stay on the GPU. Fastest, and needs the most memory.
Text encoder on CPU	13.89 GB	28 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	12.53 GB	30 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	10.17 GB	2 min 36 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

Both memory and time climb faster than the frame count, because attention is quadratic in the length of the latent sequence.

Output	Latent tokens	Peak VRAM	Time
480p, 3 seconds	305,760	16.76 GB	35 s
480p, 5 seconds	505,440	24.38 GB	58 s
720p, 3 seconds	705,600	32.02 GB	1 min 21 s
720p, 5 seconds	1,166,400	49.59 GB	2 min 15 s
720p, 10 seconds	2,318,400	93.54 GB	4 min 28 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 Video Diffusion XT

98 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	Comfortable	BF16 everything resident	13.89 GB	18 s
GeForce RTX 5090 NVIDIA	32 GB	Comfortable	BF16 everything resident	13.89 GB	21 s
RTX 6000 Ada Generation NVIDIA	48 GB	Comfortable	BF16 everything resident	13.89 GB	24 s
GeForce RTX 4090 NVIDIA	24 GB	Comfortable	BF16 everything resident	13.89 GB	27 s
GeForce RTX 5090 Laptop NVIDIA	24 GB	Comfortable	BF16 everything resident	13.89 GB	34 s
NVIDIA DGX Spark 128GB NVIDIA	128 GB	Comfortable	BF16 everything resident	13.89 GB	35 s
GeForce RTX 5080 NVIDIA	16 GB	Comfortable	BF16 everything resident	13.89 GB	39 s
GeForce RTX 4080 SUPER NVIDIA	16 GB	Comfortable	BF16 everything resident	13.89 GB	42 s
GeForce RTX 4090 Laptop NVIDIA	16 GB	Comfortable	BF16 everything resident	13.89 GB	43 s
GeForce RTX 4080 NVIDIA	16 GB	Comfortable	BF16 everything resident	13.89 GB	45 s
GeForce RTX 5080 Laptop NVIDIA	16 GB	Comfortable	BF16 everything resident	13.89 GB	46 s
GeForce RTX 5070 Ti NVIDIA	16 GB	Comfortable	BF16 everything resident	13.89 GB	49 s
GeForce RTX 4070 Ti SUPER NVIDIA	16 GB	Comfortable	BF16 everything resident	13.89 GB	49 s
GeForce RTX 4070 Ti NVIDIA	12 GB	Comfortable	GGUF Q4_K_M everything resident	11.08 GB	54 s
GeForce RTX 3090 Ti NVIDIA	24 GB	Comfortable	BF16 everything resident	13.89 GB	54 s
RTX A6000 NVIDIA	48 GB	Comfortable	BF16 everything resident	13.89 GB	56 s
GeForce RTX 4080 Laptop NVIDIA	12 GB	Comfortable	GGUF Q4_K_M everything resident	11.08 GB	58 s
GeForce RTX 3090 NVIDIA	24 GB	Workable	BF16 everything resident	13.89 GB	1 min 1 s
GeForce RTX 4070 SUPER NVIDIA	12 GB	Workable	GGUF Q4_K_M everything resident	11.08 GB	1 min 1 s
GeForce RTX 3080 Ti NVIDIA	12 GB	Workable	GGUF Q4_K_M everything resident	11.08 GB	1 min 4 s
GeForce RTX 5070 NVIDIA	12 GB	Workable	GGUF Q4_K_M everything resident	11.08 GB	1 min 10 s
GeForce RTX 3080 12GB NVIDIA	12 GB	Workable	GGUF Q4_K_M everything resident	11.08 GB	1 min 12 s
GeForce RTX 4070 NVIDIA	12 GB	Workable	GGUF Q4_K_M everything resident	11.08 GB	1 min 14 s
RTX A5000 NVIDIA	24 GB	Workable	BF16 everything resident	13.89 GB	1 min 18 s
GeForce RTX 5060 Ti 16GB NVIDIA	16 GB	Workable	BF16 everything resident	13.89 GB	1 min 30 s
GeForce RTX 4060 Ti 16GB NVIDIA	16 GB	Workable	BF16 everything resident	13.89 GB	1 min 38 s
Jetson AGX Orin 32GB NVIDIA	32 GB	Workable	BF16 everything resident	13.89 GB	1 min 41 s
Jetson AGX Orin 64GB NVIDIA	64 GB	Workable	BF16 everything resident	13.89 GB	1 min 41 s
Radeon RX 7900 XTX AMD	24 GB	Workable	BF16 everything resident	13.89 GB	1 min 42 s
Radeon PRO W7900 AMD	48 GB	Workable	BF16 everything resident	13.89 GB	1 min 43 s
RTX A4000 NVIDIA	16 GB	Workable	BF16 everything resident	13.89 GB	1 min 52 s
Radeon RX 7900 XT AMD	20 GB	Workable	BF16 everything resident	13.89 GB	2 min 2 s
Radeon RX 9070 XT AMD	16 GB	Workable	BF16 everything resident	13.89 GB	2 min 9 s
Radeon RX 7900 GRE AMD	16 GB	Workable	BF16 everything resident	13.89 GB	2 min 16 s
Radeon RX 9070 AMD	16 GB	Workable	BF16 everything resident	13.89 GB	2 min 40 s
GeForce RTX 2060 12GB NVIDIA	12 GB	Workable	GGUF Q4_K_M everything resident	11.08 GB	2 min 45 s
Radeon RX 7800 XT AMD	16 GB	Workable	BF16 everything resident	13.89 GB	2 min 47 s
GeForce RTX 3060 12GB NVIDIA	12 GB	Workable	GGUF Q4_K_M everything resident	11.08 GB	2 min 48 s
Radeon RX 7700 XT AMD	12 GB	Workable	GGUF Q4_K_M everything resident	11.08 GB	2 min 59 s
Radeon RX 6900 XT AMD	16 GB	Workable	BF16 everything resident	13.89 GB	4 min 31 s
Radeon RX 7600 XT AMD	16 GB	Workable	BF16 everything resident	13.89 GB	4 min 37 s
Ryzen AI Max+ 395 32GB AMD	32 GB	Slow	BF16 everything resident	13.89 GB	5 min 17 s
Ryzen AI Max+ 395 64GB AMD	64 GB	Slow	BF16 everything resident	13.89 GB	5 min 17 s
Ryzen AI Max+ 395 96GB AMD	96 GB	Slow	BF16 everything resident	13.89 GB	5 min 17 s
Ryzen AI Max+ 395 128GB AMD	128 GB	Slow	BF16 everything resident	13.89 GB	5 min 17 s

Source: stabilityai/stable-video-diffusion-img2vid-xt . Downloaded 0.2 million times in the last month. See how we calculate, or browse every video model.