Gemma 4 requirements: what PC you need to run it locally
Short answer: the small Gemma 4 E2B runs on an 8 GB laptop. E4B wants 16 GB of RAM, or flies on any 6 GB+ graphics card. 12B needs 10 GB+ of VRAM to hold a real conversation.
Gemma 4 requirements at a glance
| Model | File size | Minimum we'd suggest | Best on |
|---|---|---|---|
| Gemma 4 E2B | 2.6 GB | 8 GB RAM, no card | Laptops |
| Gemma 4 E4B | 4.2 GB | 16 GB RAM, or a 6 GB card | Gaming PCs, 16 GB laptops |
| Gemma 4 12B | 6.2 GB | 10 GB+ VRAM | Cards with 12 GB or more |
Measured speeds
How fast each one writes its answer, in tokens per second (a token is about three quarters of a word; you read about 5 a second). Measured with GreenCube in September and October 2026.
| PC | Model | Writes (tokens/s) |
|---|---|---|
| Desktop, AMD RX 5700 XT 8 GB | E4B | 63–72 |
| Desktop, NVIDIA RTX 5060 8 GB | E4B | 56 |
| Desktop, 12 GB graphics card | E4B | 50–55 |
| Desktop, NVIDIA GTX 1660 Super 6 GB | E4B | 35–38 |
| Desktop, AMD RX 5700 XT 8 GB | 12B | 20–35 (short chats only) |
| Desktop, NVIDIA GTX 1650 4 GB | E4B | about 20 |
| Laptop, Intel Core Ultra 5, 16 GB | E2B | 14–18 |
| Same laptop | E4B | 8–10 |
| Laptop, Intel Iris Xe, 8 GB | E2B | about 9 |
What we learned
1. E2B is the laptop model
On a 16 GB laptop with no graphics card it finished a full study answer in about 20 seconds, where E4B took 112. In five quick fact checks it got none wrong. E2B vs E4B, measured.
2. E4B is the gaming PC model
Once the whole model sits on a graphics card, E4B is 7 to 9 times faster than on a laptop, and its answers are better. Any 6 GB+ card does it.
3. 12B needs real VRAM, not just RAM
On an 8 GB card, 12B answered one short question fine, then crashed every time the chat grew. That PC had 48 GB of RAM and it didn't help. Why the model must fit on your card.
Which Gemma 4 for your PC?
8 GB laptop
E2B. Close other apps first.
16 GB laptop, no card
E2B for speed. E4B works but is slow.
Gaming PC, 6–8 GB card
E4B, fully on the card.
12 GB+ card
E4B is fast; 12B becomes an option.
Gemma 4 is released by Google under the Apache 2.0 licence. GreenCube picks the right one for your PC automatically. See all 7 PCs we measured.
Questions
What are the minimum requirements to run Gemma 4 locally?
The smallest one, Gemma 4 E2B (a 2.6 GB file), ran on a laptop with 8 GB of RAM and no graphics card, at about 9 tokens a second. It needs a couple of GB of free memory to start.
How much VRAM does Gemma 4 E4B need?
E4B is a 4.2 GB file. It fits fully on a 6 GB graphics card, where it wrote 35 to 38 tokens a second on a GTX 1660 Super. On 8 GB cards it wrote 56 to 72. It also runs without a card on 16 GB of RAM, at about 8 tokens a second.
Can Gemma 4 12B run on an 8 GB graphics card?
For one short question, yes: 20 to 35 tokens a second on an RX 5700 XT. But it crashed once the conversation got longer, because the card had almost no room left. We'd want at least 10 GB of VRAM for it.
Is Gemma 4 free to use?
Yes. Google released Gemma 4, including E2B and E4B, under the Apache 2.0 licence, which allows commercial use.
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