Home and Learn: AI Beginners Course
Check your Computer - VRAM and AI Image Creation
Created:
In the next few lessons, you'll create AI images using InvokeAI.
Along the way, you'll:
- Install InvokeAI
- Download at least one image-generation model.
- Write a simple text prompt.
- Generate your first image.
- And lots more!
Before downloading a model, however, it is worth checking how much graphics memory your computer has.
Different AI models have different hardware requirements. A smaller model may work comfortably on your computer, while a larger model may need more memory or take considerably longer to produce an image.
VRAM Explained
When you generate an image locally, much of the work is carried out by your computer's graphics processing unit, usually called the GPU.
A GPU has its own special type of memory called Dedicated Video Memory, or VRAM.
Invoke uses this memory while it is generating an image. VRAM helps hold the AI model and the information needed to construct the image.
Having more VRAM generally allows you to:
- Use larger AI models.
- Generate larger images.
- Use more demanding features.
- Avoid moving as much work into ordinary system memory.
However, more VRAM does not necessarily mean that the GPU itself is faster. Two graphics cards can have the same amount of VRAM but produce images at different speeds.
If a model needs more VRAM than your graphics card has available, Invoke may be able to move parts of the model between the GPU and your computer's ordinary RAM.
This is known as offloading. It can make a large model possible to use, but generation will usually be slower. Invoke's guidance for large models such as FLUX also warns that unnecessary model offloading can cause slow generation.
Dedicated and shared GPU memory
Windows displays several different graphics-memory figures. It is important
not to confuse them.
Dedicated GPU memory
This is the GPU's actual VRAM. This is the most important figure when
deciding what your graphics card can comfortably handle.
Shared GPU memory
This is part of your computer's ordinary RAM that Windows allows the
GPU to borrow. It is not additional VRAM and is generally slower than
dedicated GPU memory.
Total GPU memory
This combines dedicated GPU memory and the maximum amount of shared system
memory that Windows may make available.
For example, a computer with 8 GB of dedicated GPU memory and 16 GB of shared GPU memory might display 24 GB of total GPU memory. This does not mean that the computer has a 24 GB graphics card. It still has only 8 GB of dedicated VRAM.
Find your VRAM in Windows 11
To check your VRAM:
- Press Ctrl + Shift + Esc to open Task Manager.
- Select Performance from the menu on the left.
- Select GPU.
- Look for Dedicated GPU memory near the bottom of the window.
You should see something similar to this:
In this example, the computer has an NVIDIA GeForce RTX 4060 with 8.0 GB of dedicated GPU memory. Therefore, this computer has 8 GB of VRAM.
At the moment the screenshot was taken, approximately 1.6 GB of the available 8 GB was already in use.
The screenshot also shows:
- GPU memory: 1.8/24.0 GB
- Dedicated GPU memory: 1.6/8.0 GB
- Shared GPU memory: 0.1/16.0 GB
The 24 GB figure is the combined total of the computer's 8 GB of dedicated VRAM and up to 16 GB of shared system memory. It does not mean that the RTX 4060 has 24 GB of VRAM.
Don't worry if some dedicated GPU memory is already being used before you open Invoke. Windows, your web browser and other programs can all use the GPU. Closing programs that are using a large amount of GPU memory may leave more available for Invoke.
What does your result mean?
There is no single VRAM figure that guarantees every model will work. Memory use also depends on:
- The model you select.
- The width and height of the image.
- The number of images generated at once.
- Additional controls, LoRAs and other features.
- Invoke's memory-management settings.
For this reason, the safest approach is to begin with one suitable model, generate one image at a modest resolution and then increase the settings gradually.
With that out of the way, let's install InvokeAI and use it to generate an image.
Install InvokeAI, Create an Image -->