Gemma · Under 3B
Gemma 2 2B Instruct
2.61 billion parameters across 26 layers. At four-bit precision the weights alone come to 1.47 GB, before any conversation is loaded. This one is cache-hungry: a 32k window adds 1.8 GB on top, so context length matters more here than the model size suggests.
About Gemma 2 2B Instruct
Gemma 2 2B Instruct is a dense model with 2.61B parameters, built from 26 layers with a hidden size of 2304. Its repository was published on 2024-12-12. The license listed on Hugging Face is gemma. 13 of its 26 layers use a sliding window of 4k tokens, so only 13 keep growing with the conversation.
Memory needed, by quantization
At an 8k context window. Quality is our rough ranking of how much the compression costs you: anything at or above Q5 is hard to tell apart from the original in normal use.
| Quantization | File size | KV cache | Total VRAM | Quality | What it costs you |
|---|---|---|---|---|---|
| Q8_0 | 2.58 GB | 0.61 GB | 3.54 GB | 99% | Lossless in practice. Use it when the memory is there. |
| Q6_K | 1.99 GB | 0.61 GB | 2.95 GB | 98% | Very close to Q8 for two thirds of the size. |
| Q5_K_M | 1.73 GB | 0.61 GB | 2.69 GB | 96% | The quality-first choice when Q6 will not fit. |
| Q4_K_M | 1.47 GB | 0.61 GB | 2.42 GB | 93% | The default. Best size-to-quality ratio for local use. |
| IQ4_XS | 1.29 GB | 0.61 GB | 2.25 GB | 91% | Importance-matrix 4-bit. Q4_K_S quality, smaller file. |
| Q3_K_M | 1.19 GB | 0.61 GB | 2.15 GB | 86% | Degradation starts to show. A way to fit one size up. |
| IQ3_XXS | 0.93 GB | 0.61 GB | 1.89 GB | 79% | Aggressive. Only worth it on very large models. |
What a longer conversation costs
Same model at Q4_K_M, only the context window changes. 13 of its 26 layers use sliding-window attention, so their cache stops growing past 4k tokens and only the other 13 keep growing.
| Context | KV cache | Total VRAM | Fits in 8 GB | Fits in 12 GB | Fits in 24 GB |
|---|---|---|---|---|---|
| 4k | 0.41 GB | 2.22 GB | yes | yes | yes |
| 8k | 0.61 GB | 2.42 GB | yes | yes | yes |
The "fits" columns allow for the roughly 0.8 GB Windows keeps for the desktop.
Which hardware runs Gemma 2 2B Instruct
118 of 118 consumer devices run it at a quantization worth using, at an 8k context window.
| Device | Memory | Verdict | Quantization | Used | Speed |
|---|---|---|---|---|---|
| GeForce RTX 5090 NVIDIA | 32 GB | Runs great | Q8_0 | 3.54 GB | 413 t/s |
| RTX PRO 6000 Blackwell NVIDIA | 96 GB | Runs great | Q8_0 | 3.54 GB | 413 t/s |
| GeForce RTX 4090 NVIDIA | 24 GB | Runs great | Q8_0 | 3.54 GB | 232.3 t/s |
| GeForce RTX 3090 Ti NVIDIA | 24 GB | Runs great | Q8_0 | 3.54 GB | 232.3 t/s |
| GeForce RTX 5080 NVIDIA | 16 GB | Runs great | Q8_0 | 3.54 GB | 221.3 t/s |
| RTX 6000 Ada Generation NVIDIA | 48 GB | Runs great | Q8_0 | 3.54 GB | 221.3 t/s |
| GeForce RTX 3090 NVIDIA | 24 GB | Runs great | Q8_0 | 3.54 GB | 215.7 t/s |
| GeForce RTX 3080 Ti NVIDIA | 12 GB | Runs great | Q8_0 | 3.54 GB | 210.2 t/s |
| GeForce RTX 3080 12GB NVIDIA | 12 GB | Runs great | Q8_0 | 3.54 GB | 210.2 t/s |
| GeForce RTX 5070 Ti NVIDIA | 16 GB | Runs great | Q8_0 | 3.54 GB | 206.5 t/s |
| GeForce RTX 5090 Laptop NVIDIA | 24 GB | Runs great | Q8_0 | 3.54 GB | 206.5 t/s |
| Radeon RX 7900 XTX AMD | 24 GB | Runs great | Q8_0 | 3.54 GB | 187.7 t/s |
| GeForce RTX 5080 Laptop NVIDIA | 16 GB | Runs great | Q8_0 | 3.54 GB | 177 t/s |
| RTX A6000 NVIDIA | 48 GB | Runs great | Q8_0 | 3.54 GB | 177 t/s |
| RTX A5000 NVIDIA | 24 GB | Runs great | Q8_0 | 3.54 GB | 177 t/s |
| GeForce RTX 3080 10GB NVIDIA | 10 GB | Runs great | Q8_0 | 3.54 GB | 175.2 t/s |
| Apple M3 Ultra 96GB Apple | 96 GB | Runs great | Q8_0 | 3.51 GB | 174.4 t/s |
| Apple M3 Ultra 256GB Apple | 256 GB | Runs great | Q8_0 | 3.51 GB | 174.4 t/s |
| Apple M3 Ultra 512GB Apple | 512 GB | Runs great | Q8_0 | 3.51 GB | 174.4 t/s |
| Apple M1 Ultra 64GB Apple | 64 GB | Runs great | Q8_0 | 3.51 GB | 170.4 t/s |
| Apple M1 Ultra 128GB Apple | 128 GB | Runs great | Q8_0 | 3.51 GB | 170.4 t/s |
| Apple M2 Ultra 64GB Apple | 64 GB | Runs great | Q8_0 | 3.51 GB | 170.4 t/s |
| Apple M2 Ultra 128GB Apple | 128 GB | Runs great | Q8_0 | 3.51 GB | 170.4 t/s |
| Apple M2 Ultra 192GB Apple | 192 GB | Runs great | Q8_0 | 3.51 GB | 170.4 t/s |
| GeForce RTX 4080 SUPER NVIDIA | 16 GB | Runs great | Q8_0 | 3.54 GB | 169.6 t/s |
Show 35 more devices
| Radeon PRO W7900 AMD | 48 GB | Runs great | Q8_0 | 3.54 GB | 168.9 t/s |
| GeForce RTX 4080 NVIDIA | 16 GB | Runs great | Q8_0 | 3.54 GB | 165.3 t/s |
| Radeon RX 7900 XT AMD | 20 GB | Runs great | Q8_0 | 3.54 GB | 156.4 t/s |
| GeForce RTX 5070 NVIDIA | 12 GB | Runs great | Q8_0 | 3.54 GB | 154.9 t/s |
| GeForce RTX 4070 Ti SUPER NVIDIA | 16 GB | Runs great | Q8_0 | 3.54 GB | 154.9 t/s |
| GeForce RTX 2080 Ti NVIDIA | 11 GB | Runs great | Q8_0 | 3.54 GB | 142 t/s |
| GeForce RTX 3070 Ti NVIDIA | 8 GB | Runs great | Q8_0 | 3.54 GB | 140.1 t/s |
| GeForce RTX 4090 Laptop NVIDIA | 16 GB | Runs great | Q8_0 | 3.54 GB | 132.8 t/s |
| Radeon RX 9070 XT AMD | 16 GB | Runs great | Q8_0 | 3.54 GB | 126.1 t/s |
| Radeon RX 9070 AMD | 16 GB | Runs great | Q8_0 | 3.54 GB | 126.1 t/s |
| Radeon RX 7800 XT AMD | 16 GB | Runs great | Q8_0 | 3.54 GB | 122 t/s |
| Apple M4 Max 36GB Apple | 36 GB | Runs great | Q8_0 | 3.51 GB | 116.3 t/s |
| Apple M4 Max 48GB Apple | 48 GB | Runs great | Q8_0 | 3.51 GB | 116.3 t/s |
| Apple M4 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 3.51 GB | 116.3 t/s |
| Apple M4 Max 128GB Apple | 128 GB | Runs great | Q8_0 | 3.51 GB | 116.3 t/s |
| GeForce RTX 4070 Ti NVIDIA | 12 GB | Runs great | Q8_0 | 3.54 GB | 116.2 t/s |
| GeForce RTX 4070 SUPER NVIDIA | 12 GB | Runs great | Q8_0 | 3.54 GB | 116.2 t/s |
| GeForce RTX 4070 NVIDIA | 12 GB | Runs great | Q8_0 | 3.54 GB | 116.2 t/s |
| Radeon RX 7900 GRE AMD | 16 GB | Runs great | Q8_0 | 3.54 GB | 112.6 t/s |
| GeForce GTX 1080 Ti NVIDIA | 11 GB | Runs great | Q8_0 | 3.54 GB | 111.6 t/s |
| GeForce RTX 5060 Ti 16GB NVIDIA | 16 GB | Runs great | Q8_0 | 3.54 GB | 103.3 t/s |
| GeForce RTX 5060 Ti 8GB NVIDIA | 8 GB | Runs great | Q8_0 | 3.54 GB | 103.3 t/s |
| GeForce RTX 5060 NVIDIA | 8 GB | Runs great | Q8_0 | 3.54 GB | 103.3 t/s |
| GeForce RTX 3070 NVIDIA | 8 GB | Runs great | Q8_0 | 3.54 GB | 103.3 t/s |
| GeForce RTX 3060 Ti NVIDIA | 8 GB | Runs great | Q8_0 | 3.54 GB | 103.3 t/s |
| RTX A4000 NVIDIA | 16 GB | Runs great | Q8_0 | 3.54 GB | 103.3 t/s |
| Radeon RX 6900 XT AMD | 16 GB | Runs great | Q8_0 | 3.54 GB | 100.1 t/s |
| Radeon RX 6800 AMD | 16 GB | Runs great | Q8_0 | 3.54 GB | 100.1 t/s |
| Arc A770 16GB Intel | 16 GB | Runs great | Q8_0 | 3.54 GB | 99.7 t/s |
| GeForce RTX 4080 Laptop NVIDIA | 12 GB | Runs great | Q8_0 | 3.54 GB | 99.6 t/s |
| Arc A750 Intel | 8 GB | Runs great | Q8_0 | 3.54 GB | 91.1 t/s |
| Apple M1 Max 32GB Apple | 32 GB | Runs great | Q8_0 | 3.51 GB | 85.2 t/s |
| Apple M1 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 3.51 GB | 85.2 t/s |
| Apple M2 Max 32GB Apple | 32 GB | Runs great | Q8_0 | 3.51 GB | 85.2 t/s |
| Apple M2 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 3.51 GB | 85.2 t/s |
How it compares
Other models in the same family or of a similar size, on a 12 GB card and a 24 GB card, each at the best quantization that fits.
| Model | Weights at 4 bits | GeForce RTX 3060 12GB | GeForce RTX 4090 |
|---|---|---|---|
| Gemma 2 2B Instruct | 1.5 GB | 83 t/s at Q8_0 | 232.3 t/s at Q8_0 |
| Gemma 3 1B Instruct | 0.6 GB | 255 t/s at Q8_0 | 713.9 t/s at Q8_0 |
| Gemma 3 4B Instruct | 2.4 GB | 58.2 t/s at Q8_0 | 162.9 t/s at Q8_0 |
| Gemma 4 E2B Instruct | 2.9 GB | 51.7 t/s at Q8_0 | 144.9 t/s at Q8_0 |
| Qwen2.5 3B Instruct | 1.7 GB | 79.3 t/s at Q8_0 | 222.1 t/s at Q8_0 |
| Qwen3 1.7B | 1.1 GB | 91.8 t/s at Q8_0 | 257.2 t/s at Q8_0 |
What to buy to run Gemma 2 2B Instruct
The cheapest hardware that runs it at a quantization worth using and a speed you would not resent, at an 8k context window.
Cheapest that works
GeForce GTX 1650
$100 used
Q6_K · uses 2.95 GB of its 4 GB · about 36.2 tokens/s
Best value
GeForce GTX 1660 SUPER
$110 used
Q8_0 · about 77.4 tokens/s · 336 GB/s
| Device | Price | Memory | Runs it at | Speed | |
|---|---|---|---|---|---|
| GeForce GTX 1650 NVIDIA | $100 used | 4 GB | Q6_K | 36.2 t/s | Check price |
| GeForce GTX 1660 SUPER NVIDIA | $110 used | 6 GB | Q8_0 | 77.4 t/s | Check price |
| GeForce GTX 1080 Ti NVIDIA | $150 used | 11 GB | Q8_0 | 111.6 t/s | Check price |
| GeForce RTX 2060 12GB NVIDIA | $160 used | 12 GB | Q8_0 | 77.4 t/s | Check price |
| GeForce RTX 2080 Ti NVIDIA | $250 used | 11 GB | Q8_0 | 142 t/s | Check price |
| Arc A770 16GB Intel | $280 used | 16 GB | Q8_0 | 99.7 t/s | Check price |
| GeForce RTX 3080 10GB NVIDIA | $350 used | 10 GB | Q8_0 | 175.2 t/s | Check price |
| GeForce RTX 3080 12GB NVIDIA | $400 used | 12 GB | Q8_0 | 210.2 t/s | Check price |
Indicative prices reviewed 2026-09-01; used prices are marketplace typical. Options that cost more than a cheaper one with no more memory and no more speed are hidden. Price links are Amazon affiliate links. Change the standard, the context window or the budget in the buying tool.
Where it sits in the catalog
Source: Efficient-Large-Model/gemma-2-2b-it on Hugging Face. Downloaded 152k times in the last month. Published 2024-12-12. Architecture figures are read from the repository's own configuration file, so they move when the model does.
Gemma 3 1B Instruct
1B · 32k context
Gemma 3 4B Instruct
4.3B · 128k context
Gemma 4 E2B Instruct
5.12B · 128k context
Gemma 4 E4B Instruct
8B · 128k context
Gemma 2 9B Instruct
9.24B · 8k context
Gemma 4 12B Instruct
11.96B · 256k context
Gemma 3 12B Instruct
12.19B · 128k context
Gemma 4 26B A4B Instruct
25.81B (3.74B active) · 256k context
Questions people ask about Gemma 2 2B Instruct
How much VRAM does Gemma 2 2B Instruct need?
At Q4_K_M with an 8k context window, Gemma 2 2B Instruct needs about 2.42 GB: 1.47 GB of weights, 0.61 GB of KV cache and 0.35 GB of overhead. At the highest-quality file it needs about 3.54 GB.
What is the cheapest GPU that runs Gemma 2 2B Instruct?
The smallest consumer device in our catalog that fits it is the GeForce GTX 1650 with 4 GB, running it at Q6_K at roughly 36.2 tokens per second.
How much does a longer context window add for Gemma 2 2B Instruct?
On the memory side, a 32k window adds 1.8 GB of KV cache on top of the weights, and the cache grows more slowly than a conventional model because of its attention design. The longest window the model supports natively is 8k tokens.