Can my PC run AI? We measured 10 real gaming PCs and laptops
Short answer: yes. If your PC runs games, it can run AI. We measured 10 real PCs: gaming desktops from an old GTX 1650 to an RTX 2080 Ti, and two laptops with no graphics card at all. Every one of them chatted with an AI running entirely on the PC: no internet, no subscription. The only thing that changes is how fast.
The numbers
"Writes" is how fast the AI types its answer, in tokens per second. A token is about three quarters of a word. Most people read about 4 words a second (Brysbaert, 2019), roughly 5 tokens. So anything above 5 is faster than your eyes.
| PC | Graphics | RAM | AI model | Writes (tokens/s) |
|---|---|---|---|---|
| Desktop | NVIDIA RTX 2080 Ti, 11 GB | 128 GB | Gemma 4 E4B | 65–71 |
| Desktop | AMD RX 5700 XT, 8 GB | 48 GB | Gemma 4 E4B | 63–72 |
| Desktop | NVIDIA RTX 5060, 8 GB | 32 GB | Gemma 4 E4B | 56 |
| Desktop | NVIDIA RTX 3060, 12 GB | 32 GB | Gemma 4 E4B | 53–55 |
| Desktop | AMD RX 6600, 8 GB | 16 GB | Gemma 4 E4B | 42–44 |
| Desktop | NVIDIA GTX 1660 Super, 6 GB | 16 GB | Gemma 4 E4B | 37–38 |
| Desktop | NVIDIA GTX 1650, 4 GB | 16 GB | Gemma 4 E4B | about 20 |
| Laptop | Intel graphics only (Core Ultra 5) | 16 GB | Gemma 4 E2B | 14–18 |
| Desktop | AMD RX 580, 8 GB | 16 GB | Gemma 4 E4B | 10–11 |
| Laptop | Intel Iris Xe only | 8 GB | Gemma 4 E2B | 5–7 |
Click a graphics card to see its full test.
Measured with GreenCube in September and October 2026, in real chats and in GreenCube's own speed tests. Speeds move a little between runs on the same PC, so where we have several runs we show the range.
What we learned
1. Your graphics card does the heavy lifting, like in a game
On the RX 5700 XT desktop, the same AI wrote 5 tokens a second on the processor and 72 on the graphics card. AI is mostly the kind of maths graphics cards are built for. Any card from the last few years helps a lot, even an old 4 GB GTX 1650.
2. VRAM decides how big an AI fits
The AI has to fit in the card's own memory (VRAM) to run at full speed. That desktop had 48 GB of RAM, and it didn't save a bigger model: a 12-billion-parameter AI fitted on the 8 GB card for one short question, then crashed once the chat got longer, with the card almost full. For AI, VRAM beats RAM. VRAM isn't speed, though: four 8 GB cards we measured wrote anywhere from about 11 tokens a second (RX 580) to 72 (RX 5700 XT). That's why GreenCube checks your card and picks an AI that fits, not just the biggest one.
3. No graphics card? Still works
A normal 16 GB laptop with only Intel graphics wrote 14 to 18 tokens a second with the smaller AI, and finished a full answer (an explanation plus three exam questions) in about 20 seconds. Totally fine for studying and writing. An 8 GB laptop works too, at 5 to 7 tokens a second in real chats, if other apps leave it some free memory.
4. Pictures are the heavy stuff
Making a picture is way heavier than chatting. 8 to 9 seconds on an 11 GB RTX 2080 Ti (the first took 83 while it loaded), 28 seconds on a 12 GB RTX 3060, about a minute on an 8 GB RX 6600, a minute and a half on a 6 GB GTX 1660 Super, and about 2 and a half minutes on the 16 GB laptop. An 8 GB laptop can't make pictures at all: the picture AI alone needs about 7 GB.
Check your PC in 30 seconds
- Press Ctrl + Shift + Esc to open Task Manager.
- Go to Performance.
- Click Memory: the number at the top right is your RAM.
- Click GPU: the name at the top right is your card, and Dedicated GPU memory is your VRAM. Only see Intel graphics? Then you have no separate card.
What your PC will feel like
Card with 6 GB or more
Fast on most cards from the last few years: 37 to 76 tokens a second in our tests, many times quicker than you read, and pictures in seconds to a minute and a half. One exception: an old RX 580 wrote about 11, still faster than you read.
4 GB card, 16 GB RAM
Good. About 20 tokens a second in our test.
No card, 16 GB RAM
Fine for chat and study, with a smaller AI. Pictures take a couple of minutes.
No card, 8 GB RAM
Works for chat, slowly. Close other apps first. No pictures.
Why run AI on your own PC at all?
Because it's yours. No message limits, no monthly subscription, and it keeps working with no internet: on a train, on a mountain, or in a library with terrible Wi-Fi. It won't beat the biggest cloud AIs at everything, but for studying, writing, summarising and everyday questions, an AI on your own card is more than enough, and it never runs out.
Questions
Does AI on my PC work with an AMD graphics card?
Yes. We measured AMD cards (RX 5700 XT, RX 6600, RX 580) as well as NVIDIA ones. GreenCube uses Vulkan, which works on both.
How much VRAM do I need to run AI?
None is required, but 4 GB or more makes a big difference. On our test PCs, a 4 GB GTX 1650 wrote about 20 tokens a second and an 8 GB RTX 5060 about 56. More VRAM lets a bigger model fit on the card.
Can I run AI on a laptop with no graphics card?
Yes. A 16 GB laptop with only Intel graphics wrote 14 to 18 tokens a second with a small model, which is faster than you read. An 8 GB laptop works too, more slowly, and cannot make pictures.
Does more RAM make AI faster?
Not on its own. On a desktop with 48 GB of RAM, the same model ran 14 times faster once it was on the 8 GB graphics card than on the processor. RAM matters when you have no card; with a card, VRAM matters most.
Do I need internet to run AI on my PC?
Only once, to download the app and the AI model and to activate it. After that, chatting works with no internet at all.
Your own AI, on your own PC
GreenCube runs on your computer: no usage limits, no subscription, works without internet. €9.99, paid once.
Get GreenCube