Why Does AI Art Need NVIDIA? CPU vs GPU and Minimum Specs
Why does AI art need an NVIDIA graphics card, and can your PC run it? Four numbers on your spec sheet will tell you. I'm Labo, the lighthouse keeper. If you want to draw your own manga with AI at home, sooner or later someone tells you it has to be an NVIDIA card. Today the student asked me why, and the honest answer is that whether your PC can do it isn't written in any ad. Four numbers tell you.
Before anything gets installed, one look at what's inside the computer.
Last time I told her to open her Performance tab. She did. It's in Task Manager, and it shows the graphics card, its memory, and the computer's own memory on one screen. But she came in with a different question first. When she ran Kohya the other night, the black window kept printing one word: CUDA.
"Is CUDA something I turn on if I want it?"
Student: Labo, what's CUDA? Is it an option? I thought this was all supposed to run locally.
Labo: It's not an option. But let's start with why a graphics card matters at all. A drawing AI has to do thousands of calculations at once. The CPU, the brain of the computer, is clever, but it works in a line. Finish one job, start the next. The graphics card, the GPU, is more like thousands of less clever workers doing their jobs at the same time. An AI repeats the same calculation thousands of times. That's GPU work.
Not one belt, many. That's the whole difference.
Student: And CUDA?
Labo: CUDA is the passage a program uses to tell the GPU "do this calculation." NVIDIA has been building it since 2007. Kohya, SwarmUI and ComfyUI all hand their work to the graphics card through that passage. You don't choose to use it. If you have an NVIDIA card, they go through it automatically. CUDA showing up in the black window means "I'm using the graphics card right now." Things are working.
Student: So local and CUDA aren't opposites.
Labo: Right. Local means on your computer. CUDA means on the graphics card inside it.
The CPU is one line. The GPU is thousands of lines. CUDA is the passage that hands out the work to all of them.
"Does it really have to be NVIDIA?"
Student: What about AMD cards? Or a MacBook?
Labo: This part matters. For the past fifteen years, AI researchers wrote their code on top of CUDA. Kohya, SwarmUI, ComfyUI, all of it. There are ways to run them on AMD or a Mac. But you have to go around things, it's slower, and it tends to break every time something updates. It works, but it's a struggle. I don't recommend it to anyone just starting.
Student: Mine is NVIDIA. An RTX 3050.
Labo: Then you're fine. I started on a 3050 too. How much memory does the card have?
Student: 8GB. And the computer has 16GB of RAM.
"Is that enough to draw manga with AI at home?"
Labo: It is. It's exactly the minimum.
| Minimum | Comfortable | |
|---|---|---|
| Graphics card | NVIDIA, 8GB of video memory (VRAM) | 12 to 16GB |
| RAM | 16GB | 32GB |
| Disk | SSD, 50GB free | 200GB |
| Operating system | Windows 10/11 | Same |
The table is based on SDXL.
Labo: With 8GB you can generate SDXL pictures. Slowly. Teaching it your dad's style is a different job, and on 8GB of VRAM with 16GB of RAM it's right on the edge. That's the part that matters for a consistent character across panels later, because that comes from training on his drawings. If your PC gets stuck there, there's another road: renting a graphics card. RunPod, from the map. Whether you'll need it, we'll only know when you hit the wall.
"VRAM and RAM are different things?"
Labo: Different. VRAM is the memory inside the graphics card. RAM is the memory of the computer itself. The model file goes onto the VRAM, and the RAM handles the rest of the work. You have to look at both.
Labo: And here's a surprise: with 8GB of VRAM and 16GB of RAM like yours, the RAM can be the one that runs out first. I'll tell you that story when you actually get stuck on it.
Student: So before installing anything, what do I check?
Labo: Four things. Nothing else.
- Graphics card: is it NVIDIA?
- VRAM: 8GB or more?
- RAM: 16GB or more?
- SSD: 50GB free?
Labo: There are two halves to this: what you install, and what you need to have. Today was the second half. I think that order holds whether it's a desk at home or a Korean webtoon studio: first the machine, then the program.
Student: All four, yes. So now I really install things?
Labo: Now you install. You'll get stuck in three places, the same three I did, so I'll warn you before you start. And after that comes the question that matters more than any card: which of your dad's drawings to pick, and how many.
FAQ
Q. What is CUDA?
The passage programs use to hand calculations to an NVIDIA graphics card. If you have an NVIDIA card, Kohya, SwarmUI and ComfyUI use it automatically. You don't switch it on or off.
Q. Does AI art have to run on NVIDIA?
Not strictly, but in practice yes. The programs were all built on CUDA, so on AMD or a Mac you have to work around things, it's slower, and it breaks more often.
Q. Can I draw manga with AI at home on an 8GB graphics card?
Generating, yes, slowly. Training your own style also depends on your RAM, and if your PC can't handle it you can rent a GPU on RunPod. To check your numbers on Windows, open Task Manager and look at the Performance tab.
What the artist did: checked four numbers on her own computer (NVIDIA, 8GB VRAM, 16GB RAM, SSD) before installing anything.
What the AI did: nothing yet. CUDA only showed up in the black window to say the graphics card was being used.
Official pages: NVIDIA CUDA (you usually don't need to install it separately; it comes with the driver), NVIDIA drivers (keep these up to date), and RunPod (rent a graphics card when your home PC isn't enough, paid).
Written October 2026. The spec table is based on SDXL; screens and versions may change. The original Korean post: AI 그림은 왜 엔비디아예요? (Korean)
Lighthouse log. Whether a ship can make the crossing isn't in the brochure. Four numbers say it: graphics card, VRAM, RAM, SSD.
Keep reading
- Previous post: What Is Stable Diffusion? From a Company Server to Your PC
- More on beginner · GPU
- All posts in Draw Manga with AI at Home


Comments
Post a Comment