How much VRAM does Stable Audio Open need?

At BF16 with everything resident, Stable Audio Open peaks at about 3.86 GB. Offloading components one at a time brings that to 3.66 GB, and quantising the weights to 4-bit brings it to 2.29 GB.

What is the cheapest GPU that runs Stable Audio Open?

The GeForce GTX 1650 at around $100, running it at GGUF Q8_0 with everything resident, taking about 10 min 1 s per generation.

Stable Audio · Diffusion transformer · Music and sound

Stable Audio Open

Stable Audio Open is a pipeline of 3 networks totalling 1.323 billion parameters, of which 1.057 billion do the actual generating. It produces the whole utterance in parallel rather than token by token, which is why models this small can run many times faster than real time.

Whole pipeline1.323B

Denoiser1.057B

Download4.9 GB

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	1.057B	F32	3.94 GB	Does the generating. Runs once per step, so it dominates both memory and time.
text_encoder	0.11B	F32	0.41 GB	Turns your prompt into conditioning. Runs once, then can leave the GPU entirely.
vae	0.156B	F32	0.58 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	3.86 GB	100%	How the weights are published. No loss, and the largest footprint.
FP8	2.78 GB	97%	Halves the denoiser with a small, usually invisible cost. Needs Ada or newer.
GGUF Q8_0	2.85 GB	98%	Works on any card, unlike FP8. Slightly slower than native precision.
GGUF Q5_K_M	2.46 GB	95%	A middle step when Q8 will not fit.
GGUF Q4_K_M	2.35 GB	91%	The usual way a 12B image model gets onto an 8 GB card. Detail softens.
NF4	2.29 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 80% of the pipeline and stays resident whatever you rearrange.

Arrangement	Peak VRAM	Time	How it works
Everything resident	3.86 GB	11 s	All components stay on the GPU. Fastest, and needs the most memory.
Text encoder on CPU	3.66 GB	11 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	3.37 GB	11 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	1.67 GB	11 s	Weights stream layer by layer from system RAM. Runs almost anything on almost anything, and is many times slower.

Which hardware runs Stable Audio Open

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	3.86 GB	8.5x real time
GeForce RTX 5090 NVIDIA	32 GB	Workable	BF16 everything resident	3.86 GB	7.1x real time
RTX 6000 Ada Generation NVIDIA	48 GB	Workable	BF16 everything resident	3.86 GB	6.2x real time
GeForce RTX 4090 NVIDIA	24 GB	Workable	BF16 everything resident	3.86 GB	5.6x real time
GeForce RTX 5090 Laptop NVIDIA	24 GB	Workable	BF16 everything resident	3.86 GB	4.4x real time
NVIDIA DGX Spark 128GB NVIDIA	128 GB	Workable	BF16 everything resident	3.86 GB	4.2x real time
GeForce RTX 5080 NVIDIA	16 GB	Workable	BF16 everything resident	3.86 GB	3.8x real time
GeForce RTX 4080 SUPER NVIDIA	16 GB	Workable	BF16 everything resident	3.86 GB	3.5x real time
GeForce RTX 4090 Laptop NVIDIA	16 GB	Workable	BF16 everything resident	3.86 GB	3.4x real time
GeForce RTX 4080 NVIDIA	16 GB	Workable	BF16 everything resident	3.86 GB	3.3x real time
GeForce RTX 5080 Laptop NVIDIA	16 GB	Workable	BF16 everything resident	3.86 GB	3.2x real time
GeForce RTX 5070 Ti NVIDIA	16 GB	Workable	BF16 everything resident	3.86 GB	3x real time
GeForce RTX 4070 Ti SUPER NVIDIA	16 GB	Workable	BF16 everything resident	3.86 GB	3x real time
GeForce RTX 4070 Ti NVIDIA	12 GB	Workable	BF16 everything resident	3.86 GB	2.7x real time
GeForce RTX 3090 Ti NVIDIA	24 GB	Workable	BF16 everything resident	3.86 GB	2.7x real time
RTX A6000 NVIDIA	48 GB	Workable	BF16 everything resident	3.86 GB	2.6x real time
GeForce RTX 4080 Laptop NVIDIA	12 GB	Workable	BF16 everything resident	3.86 GB	2.5x real time
GeForce RTX 3090 NVIDIA	24 GB	Workable	BF16 everything resident	3.86 GB	2.4x real time
GeForce RTX 4070 SUPER NVIDIA	12 GB	Workable	BF16 everything resident	3.86 GB	2.4x real time
GeForce RTX 3080 Ti NVIDIA	12 GB	Workable	BF16 everything resident	3.86 GB	2.3x real time
GeForce RTX 5070 NVIDIA	12 GB	Workable	BF16 everything resident	3.86 GB	2.1x real time
Radeon RX 7900 XTX AMD	24 GB	Workable	BF16 everything resident	3.86 GB	2.1x real time
Radeon PRO W7900 AMD	48 GB	Workable	BF16 everything resident	3.86 GB	2.1x real time
GeForce RTX 3080 12GB NVIDIA	12 GB	Workable	BF16 everything resident	3.86 GB	2x real time
GeForce RTX 3080 10GB NVIDIA	10 GB	Workable	BF16 everything resident	3.86 GB	2x real time
GeForce RTX 4070 NVIDIA	12 GB	Slow	BF16 everything resident	3.86 GB	2x real time
RTX A5000 NVIDIA	24 GB	Slow	BF16 everything resident	3.86 GB	1.9x real time
GeForce RTX 2080 Ti NVIDIA	11 GB	Slow	BF16 everything resident	3.86 GB	1.8x real time
Radeon RX 7900 XT AMD	20 GB	Slow	BF16 everything resident	3.86 GB	1.7x real time
Radeon RX 9070 XT AMD	16 GB	Slow	BF16 everything resident	3.86 GB	1.6x real time
GeForce RTX 5060 Ti 16GB NVIDIA	16 GB	Slow	BF16 everything resident	3.86 GB	1.6x real time
GeForce RTX 5060 Ti 8GB NVIDIA	8 GB	Slow	BF16 everything resident	3.86 GB	1.6x real time
GeForce RTX 4070 Laptop NVIDIA	8 GB	Slow	BF16 everything resident	3.86 GB	1.6x real time
Radeon RX 7900 GRE AMD	16 GB	Slow	BF16 everything resident	3.86 GB	1.6x real time
GeForce RTX 4060 Ti 16GB NVIDIA	16 GB	Slow	BF16 everything resident	3.86 GB	1.5x real time
GeForce RTX 4060 Ti 8GB NVIDIA	8 GB	Slow	BF16 everything resident	3.86 GB	1.5x real time
GeForce RTX 3070 Ti NVIDIA	8 GB	Slow	BF16 everything resident	3.86 GB	1.5x real time
Jetson AGX Orin 32GB NVIDIA	32 GB	Slow	BF16 everything resident	3.86 GB	1.4x real time
Jetson AGX Orin 64GB NVIDIA	64 GB	Slow	BF16 everything resident	3.86 GB	1.4x real time
GeForce RTX 3070 NVIDIA	8 GB	Slow	BF16 everything resident	3.86 GB	1.4x real time
Radeon RX 9070 AMD	16 GB	Slow	BF16 everything resident	3.86 GB	1.3x real time
GeForce RTX 5060 NVIDIA	8 GB	Slow	BF16 everything resident	3.86 GB	1.3x real time
RTX A4000 NVIDIA	16 GB	Slow	BF16 everything resident	3.86 GB	1.3x real time
Radeon RX 7800 XT AMD	16 GB	Slow	BF16 everything resident	3.86 GB	1.3x real time
Radeon RX 7700 XT AMD	12 GB	Slow	BF16 everything resident	3.86 GB	1.2x real time

Source: stabilityai/stable-audio-open-1.0 . Downloaded 0.0 million times in the last month. See how we calculate, or browse every audio model.