How To Install & Use SUPIR : SOTA Image Upscaler On RunPod - 1 Click Easy Install & Run
SUPIR is still the unchallenged image upscaler; the V91 update brings Torch 2.8, CUDA 12.9, a resumable hash-checked model downloader and faster installs.
Published Apr 13, 2024 · 4,160 views · Watch on YouTube
Get the app and files

SUPIR Image Upscaler Update: RTX 1000 to RTX 5000 & Cloud GPU Support
SUPIR is still the unchallenged image upscaler; the V91 update brings Torch 2.8, CUDA 12.9, a resumable hash-checked model downloader and faster installs.

SUPIR Upscaler 1-Click Installer & Free Kaggle Notebook (Legacy)
SUPIR upscaler 1-click installers for Windows, RunPod, Massed Compute and Linux plus a free Kaggle notebook; this older copy points to the maintained post.
Chapters
- 0:00 Introduction and Setup
- 0:48 RunPod Deployment
- 1:36 Installation and Troubleshooting
- 3:24 Re-deploying for Performance
- 5:34 Launching the Upscaler
- 6:43 Using the User Interface
- 7:33 Image Processing and Testing
About this video
Download installation scripts here : https://www.patreon.com/posts/99176057
Register RunPod here : https://runpod.io?ref=1aka98lq
Hello everyone, in this video I am going to show you how to install our custom-designed, the very best upscaler right now, the SUPIR
installation on RunPod on a very cheap GPU, and I will explain to you the likely errors that you are going to encounter so you will not have any issues while you are installing our application on RunPod. You see, currently, this is 3090 GPU and only 26 cents per hour, it is super cheap and it is working amazingly. You see, this is 2 megapixels upscale and I can upscale up to 5, 6, 8 megapixels on this GPU, maybe even more. So let's begin. Okay guys, in this video I am going to show you how to install SUPIR on RunPod quickly because some people are having problems. Download zip file, then let's go to our RunPod, log in your RunPod, click Pod Deploy GPU Pod, I'm going to deploy onto Community Cloud, select extreme speed here. Now you can pick any GPU that has over 24 GB. For this tutorial, I will show RTX 3090, but you can also get A6000 GPU, so it is up to you. Let's get 3090, you can also click Change template and type PyTorch to get it like this, edit Pod, make it volume disk like at least 80 GB like this, set override and Deploy on demand. Now go to my pods, here there is a very important thing that you have to be careful of, make sure that you Pod has a minimum of 31 GB. In some cases, it is giving you 20 GB RAM memory and if you get such RAM memory, it will not work. Make sure that your Pod has over 31 GB RAM memory otherwise it may not work accurately. So this pod has 31 GB, therefore it should work very well. After the connect button appeared, click connect, connect to Jupyter lab. Okay, it's not ready yet, so you need to wait until it becomes ready. Let's refresh. Okay, not ready yet, let's look at the logs.
Okay yeah, it became ready, so it may take a while. This Ram is extremely important, also if you want more speed, you should get A6000 GPU. I will show A6000 on Massed Compute because it will be better for you to use on Massed Compute, it is much cheaper. So let's upload all the extracted files into there, then open the RunPod instructions, by the way, instructions are the same for every GPU, it doesn't matter. Okay, open a terminal, copy-paste this command, this is for installation. Now wait and look at the messages carefully to see whether there are any errors or not. You see, this Pod is very slow, you see the download speed, therefore I will not wait and I will get a new Pod, and also there were some yellow icon here, so if your Pod is slow like this, do not wait, just delete it and get a new pod. So I will get a new pod because there are so many pods on the RunPod which are not good to use, unfortunately. I will also this time filter for NVMe SSD disk so it should be better. Okay, everything is set, let's set the volume disk one more time, Deploy on demand. Let's go to My Pods, okay this one has 23 GB so it will not work, so let's delete this one too, unfortunately, this is something that you have to try. Let's try again, so let's make the system gigabyte per GPU like 48 GB, yes and let's go 3090, okay, edit template, let's make this 80 and Deploy on demand, My Pods. Okay this time I have an amazing Pod, the speed not looking good for the hard drive but it has 77 GB Ram, let's see what we will get. Okay, new GPU Pod started, let's connect to Jupyterlab and see the speed of this one, the previous one was very bad. I'm not going to delete this parts of the video because these are the most common errors that you encounter and then think it doesn't work because of the RunPod system. I'm going to introduce you Massed Compute which is much better, hopefully, I will make a tutorial for that. Okay, it is starting, let's see, this one looks faster than the previous one, yeah the speed of this Pod is many times better, many many times better so it was a good choice to delete the previous one and load this one. This one has over 100 megabytes per second, previous one were struggling with 2 megabytes per second.
You see, getting a fast Pod is super important, it is downloading models with 90 MB per second. If you can find a faster Pod your installation will be faster. Alright, everything installed, you see, all the models are downloaded also verify that you don't have major errors, this is not important, we fix that, and you see, all the messages are here so we are ready. Then few options to start the SUPIR gradio. Okay, since this is a 24 GB GPU, I'm going to start as fp8. I'm going to use tiling, tiling is really slowing down the speed, but it allows you to generate much higher resolution so I am going to start with using tiled one.