How to run Bark text-to-audio on a cloud GPU
Rent a GPU on Deploy, then open a terminal in JupyterLab or over SSH. Install Bark with the command the suno-ai/bark README prints, run the Python example that writes bark_generation.wav, and copy that file off the instance before you destroy it. Destroying the instance deletes its disk.
Other speech and audio pages: How to run XTTS-v2 on a cloud GPU, How to run Kokoro TTS on a cloud GPU, How to run Orpheus TTS on a cloud GPU, How to run Chatterbox TTS on a cloud GPU, How to run Stable Audio Open on a cloud GPU, and How to run Spark-TTS on a cloud GPU. The Spark-TTS README prints no GB figure. If the process runs out of memory: How to fix CUDA out of memory on a cloud GPU.
The README says Bark is licensed under the MIT License, and a 2023.05.01 update says it is available for commercial use. The LICENSE file is the MIT License. Its copyright line is Copyright © Suno, Inc. pyproject.toml contains a comment that says Apache 2.0. The GitHub license field on 2026-11-13 was MIT. The suno/bark model card front matter also says license: mit. This page does not merge the pyproject.toml comment with the LICENSE file. Read the LICENSE file before you redistribute the program.
The commands below are copied from the upstream files named in each step. They were not run on a CloudGPU instance for this page. Anything this page adds around them is marked.
What the README says about memory
The README hardware section says Bark has been tested on CPU and GPU (pytorch 2.0+, CUDA 11.7 and CUDA 12.0). It says the full version of Bark requires around 12GB of VRAM to hold everything on GPU at the same time, and that SUNO_USE_SMALL_MODELS=True should fit into 8GB VRAM. The FAQ on the same page says the full version requires around 12Gb of memory, and that smaller cards down to ~2Gb work with this snippet, which the FAQ says to add before generation:
import os
os.environ["SUNO_OFFLOAD_CPU"] = "True"
os.environ["SUNO_USE_SMALL_MODELS"] = "True"
The 2023.05.01 update says you can use Bark with GPUs that have low VRAM (<4GB). This page does not collapse 12GB and 12Gb, and it does not collapse <4GB and ~2Gb. The 8GB sentence in the hardware section sets only SUNO_USE_SMALL_MODELS. The FAQ snippet sets that flag and SUNO_OFFLOAD_CPU. bark/generation.py reads SUNO_USE_SMALL_MODELS and SUNO_OFFLOAD_CPU, both defaulting to False, and it also reads SUNO_ENABLE_MPS, default False. The README snippet above does not set SUNO_ENABLE_MPS.
The same update says 2x speed-up on GPU and 10x speed-up on CPU. The hardware section says that on enterprise GPUs and PyTorch nightly, Bark can generate audio in roughly real-time, and that on older GPUs, default colab, or CPU, inference time might be significantly slower. This page did not measure those sentences.
GET /v1/public/compute-capabilities does not list GPUs. It returns feature flags. The memory column is vram_gb from gpu-prices.json, updated_at 2026-11-13T02:10:27.641Z. GET /v1/public/gpu-catalog the same day (generated_at 2026-11-13T02:10:28.744Z) listed the same names and the same vram_gb values, plus Ascend 910B at 64, which is not in gpu-prices.json and is not listed here. The catalog response also includes an availability field. This page does not copy it. Every prices row has gpu_count 1. Every vram_gb below is 8 or higher. Rows at 12 or higher are the ones whose printed vram_gb is at least the 12 in the 12GB and 12Gb sentences. The 8GB sentence meets every row as a printed number. No row is under 4, so the <4GB sentence and the ~2Gb sentence do not name a card in this table. This page did not run Bark and does not mark a row as enough. Which listing you can start is on Deploy.
The catalog template list had no Bark template. The ComfyUI examples top-level directory list on 2026-11-13 had no directory named for this model. The audio example README in that repository does not name Bark. Examples HEAD was f9431bb000ce792094ff345446e22cac1ea6cef3, committed 2025-11-26T20:33:58Z. This page does not add a ComfyUI install.
| Catalog name | vram_gb |
|---|---|
| RTX 2080 | 8 |
| RTX 3080 | 10 |
| RTX 2080 Ti | 11 |
| RTX 3080 Ti | 12 |
| V100 16G | 16 |
| A30 | 24 |
| RTX 3090 | 24 |
| RTX 4090 | 24 |
| RTX 4090D | 24 |
| RTX 4080 SUPER 32G | 32 |
| RTX 5090 | 32 |
| RTX 5090D | 32 |
| V100 32G | 32 |
| A100 40G | 40 |
| A800 40G | 40 |
| A40 | 48 |
| RTX 4090 48G | 48 |
| RTX 4090D 48G | 48 |
| RTX A6000 | 48 |
| NVIDIA RTX PRO 5000 72G | 72 |
| RTX PRO 5000 72G | 72 |
| A100 80G | 80 |
| A800 80G | 80 |
| H800 | 80 |
| RTX 6000D | 83 |
| H20 96G | 96 |
| RTX PRO 6000 96G | 96 |
| RTX PRO 6000 Max-Q | 96 |
| H20-3e 141G | 141 |
Install and run the documented example
On Deploy, choose Raw GPU. The deploy page says “SSH access and root within a dedicated container”. The README does not print a Python version. pyproject.toml says requires-python >=3.8.
The installation section says: do not use pip install bark. It says that command installs a different package, which is not managed by Suno. The command it prints is:
pip install git+https://github.com/suno-ai/bark.git
or:
git clone https://github.com/suno-ai/bark
cd bark && pip install .
PyPI package bark 0.1.5, read 2026-11-13, had project_urls.source https://github.com/suno-ai/bark and the same dependency names as pyproject.toml. This page does not treat that metadata as permission to ignore the README caution, and it did not install either command.
The commits API for the default branch main listed f4f32d4cd480dfec1c245d258174bc9bde3c2148, committed 2024-04-05T13:14:16Z, message “Update README.md”. The repository object the same day said pushed_at 2024-08-19T07:45:36Z. This page did not install the package.
The README basics example is:
from bark import SAMPLE_RATE, generate_audio, preload_models
from scipy.io.wavfile import write as write_wav
from IPython.display import Audio
# download and load all models
preload_models()
# generate audio from text
text_prompt = """
Hello, my name is Suno. And, uh — and I like pizza. [laughs]
But I also have other interests such as playing tic tac toe.
"""
audio_array = generate_audio(text_prompt)
# save audio to disk
write_wav("bark_generation.wav", SAMPLE_RATE, audio_array)
# play text in notebook
Audio(audio_array, rate=SAMPLE_RATE)
bark/generation.py sets SAMPLE_RATE = 24_000. The small-model checkpoint names in that file are text.pt, coarse.pt, and fine.pt. The other three names are text_2.pt, coarse_2.pt, and fine_2.pt. Every repo_id in that map is suno/bark. This page did not download those files.
The README says generate_audio works well with around 13 seconds of spoken text. The FAQ asks why output is limited to ~13-14 seconds and says the architecture is optimized to output generations with roughly this length. This page does not collapse those two lengths. Longer audio is the notebook notebooks/long_form_generation.ipynb. This page does not paste that notebook.
The command line the README prints is:
python -m bark --text "Hello, my name is Suno." --output_filename "example.wav"
bark/cli.py defaults --output_filename to bark_generation.wav and --output_dir to .. It joins those two and prints Done! Output audio file is saved at: and then the path. This page did not run the module.
Speaker prompts
The README voice-preset example passes history_prompt="v2/en_speaker_1":
text_prompt = """
I have a silky smooth voice, and today I will tell you about
the exercise regimen of the common sloth.
"""
audio_array = generate_audio(text_prompt, history_prompt="v2/en_speaker_1")
The README says Bark supports 100+ speaker presets across the supported languages, that you can browse them in the linked library or in bark/assets/prompts, and that Bark does not currently support custom voice cloning. bark/assets/prompts/readme.md says the v2 prompts are the ones to use by including v2 in the prompt, and it repeats history_prompt="v2/en_speaker_1". cli.py defaults --history_prompt to None. Its help text is history choice for audio cloning, be path to the .npz file. This page does not treat that help text and the README cloning sentence as the same claim.
The supported-language table marks English (en), German (de), Spanish (es), French (fr), Hindi (hi), Italian (it), Japanese (ja), Korean (ko), Polish (pl), Portuguese (pt), Russian (ru), Turkish (tr), and Chinese, simplified (zh).
The Transformers section says Bark is available in Transformers from version 4.31.0 onwards. The install line it prints is pip install git+https://github.com/huggingface/transformers.git. The Python block loads suno/bark and sets voice_preset = "v2/en_speaker_6". The scipy block in that section writes bark_out.wav and takes the rate from model.generation_config.sample_rate. The model card README also has a block that writes /path/to/audio.wav and a block that reads model.config.sample_rate. This page does not collapse those rates with SAMPLE_RATE in generation.py, and it did not run the Transformers example.
No local web UI in this repository
The README does not print a Gradio launch line. The repository file list has no Gradio script. A badge links a hosted Space. That is not a command this page copies. There is no port to forward. How to use SSH port forwarding on a cloud GPU is the tunnel page for programs that do open a port.
Where the audio is saved
The basics example writes bark_generation.wav. The README command line passes --output_filename "example.wav". cli.py defaults the filename to bark_generation.wav and the directory to .. The Transformers scipy example writes bark_out.wav. The model card README example writes /path/to/audio.wav. Copy the file the command you ran actually wrote. This page did not run any of them.
When it fails
CUDA out of memory. The documented memory settings are the 12GB sentence, the 8GB SUNO_USE_SMALL_MODELS sentence, the <4GB update, and the FAQ snippet that sets SUNO_OFFLOAD_CPU and SUNO_USE_SMALL_MODELS. The README does not print an out-of-memory flag. If the process raises that error, use how to fix CUDA out of memory on a cloud GPU.
“CUDA error: no kernel image is available for execution on the device” on an RTX 5090. The README names pytorch 2.0+, CUDA 11.7, and CUDA 12.0. It does not name compute capability 12.0. The RTX 5090 guide says that card is compute capability 12.0 (sm_120) and that PyTorch 2.7 and later, on a build for CUDA 12.8 or newer, is the line that announced support. This page did not install a torch wheel and did not re-test this package on an RTX 5090. Catalog RTX 5090 and RTX 5090D are 32. That is the catalog number only.
Save the audio, then destroy
Copy bark_generation.wav, or the path the command you ran actually wrote, off the instance before you destroy it. Save your work before you destroy a cloud GPU is the checklist. The terms say data on local instance storage is permanently deleted when the instance is destroyed.
The pricing page FAQ states the billing rule in these words:
Per-minute options have a 1-minute minimum; unused prepaid minutes are refunded when you destroy the instance. Some configurations use whole-hour billing with a 1-hour minimum; unused time within a charged hour is not refunded on voluntary stop. The billing period, minimum charge and initial prepayment are shown before confirmation. No subscription.
Prices are not copied here.
Frequently asked questions
How much VRAM does the Bark README print?
The hardware section says the full version requires around 12GB of VRAM, and that SUNO_USE_SMALL_MODELS=True should fit into 8GB VRAM. The FAQ on that README says around 12Gb of memory, and smaller cards down to ~2Gb when SUNO_OFFLOAD_CPU and SUNO_USE_SMALL_MODELS are set. A 2023.05.01 update says low VRAM (<4GB). This page does not collapse those figures. Every vram_gb in https://cloudgpu.app/gpu-prices.json (updated_at 2026-11-13T02:10:27.641Z) is 8 or higher. No listed row is under 4. This page did not run Bark and does not mark a row as enough. Which listing you can start is on https://cloudgpu.app/deploy.
Where does the documented example save audio, and which speaker string does it pass?
The Python example calls write_wav with bark_generation.wav and SAMPLE_RATE. bark/generation.py sets SAMPLE_RATE to 24000. The command line example passes --output_filename example.wav. cli.py defaults --output_filename to bark_generation.wav and --output_dir to the current directory. The voice preset example passes history_prompt v2/en_speaker_1. The Transformers example passes voice_preset v2/en_speaker_6 and writes bark_out.wav. The README says Bark does not currently support custom voice cloning. The cli.py help text is: history choice for audio cloning, be path to the .npz file. This page did not run either example.
Does the README ship a local web UI or a ComfyUI workflow?
The README does not print a Gradio launch command. The repository file list has no Gradio script. The ComfyUI examples top-level directory list on 2026-11-13 had no Bark directory. Examples HEAD was f9431bb000ce792094ff345446e22cac1ea6cef3, committed 2025-11-26. The catalog template list had no Bark template. This page does not add a ComfyUI install and does not add a tunnel.
What license is on the repository?
The README says Bark is licensed under the MIT License. The LICENSE file is the MIT License, and the GitHub license field was MIT. pyproject.toml contains a comment that says Apache 2.0. This page does not merge that comment with the LICENSE file. The LICENSE grant includes the words free of charge, and that phrase is license text, not a CloudGPU price and not a storage allowance.
Sources (read 2026-11-13, Asia/Shanghai)
- suno-ai/bark README (install caution,
pip install git+https://github.com/suno-ai/bark.git, the clone line, the Python example,history_prompt="v2/en_speaker_1", the command line, 12GB, 8GB,<4GB, the FAQ 12Gb and ~2Gb snippet, MIT, around 13 seconds, ~13-14 seconds, 100+ presets, the language table, the Transformers block): https://github.com/suno-ai/bark/blob/main/README.md commitf4f32d4cd480dfec1c245d258174bc9bde3c2148 bark/generation.py(SAMPLE_RATE = 24_000,SUNO_USE_SMALL_MODELS,SUNO_OFFLOAD_CPU,SUNO_ENABLE_MPS, checkpoint file names,repo_idsuno/bark) andbark/cli.py(output defaults and help text): that commitbark/assets/prompts/readme.md(v2/en_speaker_1): that commitLICENSE(MIT, Copyright © Suno, Inc). GitHub license field MIT.pyproject.tomlcomment says Apache 2.0.requires-python>=3.8- suno/bark model card (
license: mit,gatedfalse,lastModified2023-10-04T14:17:55.000Z, no GB line,/path/to/audio.wav,model.config.sample_rate): https://huggingface.co/suno/bark - PyPI
bark0.1.5 (project_urls.sourcehttps://github.com/suno-ai/bark). This page did not install it - ComfyUI examples top-level directories and
audio/README.md(no Bark): https://github.com/comfyanonymous/ComfyUI_examples commitf9431bb000ce792094ff345446e22cac1ea6cef3 - Public GPU memory: https://cloudgpu.app/gpu-prices.json (
updated_at2026-11-13T02:10:27.641Z) and https://cloudgpu.app/v1/public/gpu-catalog (generated_at2026-11-13T02:10:28.744Z).GET /v1/public/compute-capabilitiesreturned feature flags and no GPU list.
Cards you can start are on the Deploy page. Prices are on https://cloudgpu.app/pricing and are not repeated here.
Last updated: 2026-11-13 (Asia/Shanghai).