24GB VRAM for $0.22 ?- Cheap GPUs for ComfyUI and FLUX | Easy 1 click installation and downloads

Introduction

When working with advanced machine learning models, having high GPU specifications can be crucial for efficiency and performance. Recently, I faced a similar challenge while experimenting with negative prompting techniques for the FLUX model, which required a substantial GPU setup, specifically around 24GB VRAM. However, my local machine fell short in that regard. Here’s how I navigated this situation with an easy one-click solution to access powerful cloud GPUs.

Accessing Cloud GPUs

I found a service that provides cheap cloud GPUs where you can rent servers equipped with GPUs ranging from 24GB VRAM for just $ 0.22 per hour. This isn’t a subscription service; you pay based on your actual usage. For instance, if you utilize a GPU for only 15 minutes, you’ll only be charged for that short period.

Setting Up Your Cloud Environment

To get started, you’ll need to create an account on the platform (let's refer to it as Runpod). Think of this platform as a wallet where you can add credits via your credit card. A deposit of $ 10 will remain available as you use the GPU, without any ongoing costs.

Explore the Environment: Once logged in, navigate to the "Explore" section and search for "ComfyUI." You’ll see various Docker containers shared by the community. Select the ComfyUI option.
Deploy the GPU Pod: Click on "Deploy," ensuring you select the “Community Cloud” for a more affordable rate. Choose a configuration that offers at least 24GB VRAM, such as the RDX 390 or an A30, both priced at $ 0.22 per hour.
Modify the Volume Disk Space: Adjust the disk space from the default 80GB to just 20GB to minimize costs.
Start Your Pod: Once the setup is complete, click on "Connect" to access your working environment.

Using ComfyUI

After clicking connect, you’ll have access to multiple ports. The critical ports for your work are for ComfyUI (port 3000) and Jupyter Lab (port 8888). The Jupyter Lab is useful for loading files, while ComfyUI allows you to manage the workflows involving models and setups.

Uploading the Setup Script

As you proceed, upload the pre-prepared script for automated installation of the necessary AI models and nodes. Once uploaded, activate it to initiate the download process for everything you need, including the FLUX model, text encoders, and custom nodes.

Monitoring Your Progress

The installation process will download various models automatically, tracking the progress on the status bar. Depending on your internet speed and the weight of the files (for example, the FLUX models at about 9.18GB), this could take some time.

Updating and Managing Your Pod

After successful installation, update ComfyUI through the manager. Occasionally, some custom nodes may fail to load. If that happens, you can fix any missing nodes using the “Install Missing Custom Nodes” feature. Always remember to stop the Pod after use to avoid ongoing charges.

On cessation, you’ll retain your setup in the cloud storage for future use, allowing for a potentially very cost-effective experience.

Saving Generated Outputs

For any images or data generated during your session, you’ll find them in the "outputs" folder in ComfyUI. You can download them easily to your local machine, thus ensuring all your work is preserved.

Conclusion

Once your work is complete, remember to stop the Pod to prevent incurring further charges. You can always terminate the Pod and release the storage when you’re entirely done. The whole setup process is fairly simple, and with this one-click solution for cloud GPUs, you can run high-demand workflows without the need for expensive local hardware.

Keywords

Cloud GPU
ComfyUI
FLUX Model
24GB VRAM
Runpod
Jupyter Lab
Custom Nodes
One-click installation
Cost-effective

FAQ

Q1: How much does renting a cloud GPU cost?
A1: The cost starts at approximately $ 0.22 per hour for a GPU with 24GB VRAM.

Q2: Do I have to pay a monthly subscription?
A2: No, you only pay for the time you actually use the GPU.

Q3: How can I access ComfyUI and other tools?
A3: After launching your Pod, you can connect to various ports, with ComfyUI typically on port 3000.

Q4: Can I save and download my outputs?
A4: Yes, you can save generated images and data to a designated outputs folder and download them to your local storage.

Q5: What happens to my Pod when I stop it?
A5: Stopping your Pod halts GPU usage charges, but any data and configurations are saved in cloud storage for future access.