Gemma · 9B to 32B
Gemma 2 9B Instruct
9.24 billion parameters across 42 layers. At four-bit precision the weights alone come to 5.20 GB, before any conversation is loaded. Grouped-query attention keeps the cache small, so long contexts cost less here than on models of the same size.
Memory needed, by quantisation
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.
| Quantisation | File size | KV cache | Total VRAM | Quality | What it costs you |
|---|---|---|---|---|---|
| Q8_0 | 9.14 GB | 1.31 GB | 11.28 GB | 99% | Lossless in practice. Use it when the memory is there. |
| Q6_K | 7.06 GB | 1.31 GB | 9.19 GB | 98% | Very close to Q8 for two thirds of the size. |
| Q5_K_M | 6.12 GB | 1.31 GB | 8.25 GB | 96% | The quality-first choice when Q6 will not fit. |
| Q4_K_M | 5.2 GB | 1.31 GB | 7.33 GB | 93% | The default. Best size-to-quality ratio for local use. |
| IQ4_XS | 4.57 GB | 1.31 GB | 6.7 GB | 91% | Importance-matrix 4-bit. Q4_K_S quality, smaller file. |
| Q3_K_M | 4.21 GB | 1.31 GB | 6.34 GB | 86% | Degradation starts to show. A way to fit one size up. |
| IQ3_XXS | 3.29 GB | 1.31 GB | 5.42 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. This model uses sliding-window attention, so most layers stop growing past 4k tokens and the bill flattens out.
| Context | KV cache | Total VRAM | Fits in 8 GB | Fits in 12 GB | Fits in 24 GB |
|---|---|---|---|---|---|
| 4k | 1.31 GB | 7.22 GB | no | yes | yes |
| 8k | 1.31 GB | 7.33 GB | no | yes | yes |
The "fits" columns allow for the roughly 0.8 GB Windows keeps for the desktop.
Which hardware runs Gemma 2 9B Instruct
111 of 118 consumer devices run it at a quantisation worth using, at an 8k context window. Another 6 can load it only by compressing the weights far enough to damage the model, marked with a warning below.
| Device | Memory | Verdict | Quantisation | Used | Speed |
|---|---|---|---|---|---|
| GeForce RTX 5090 NVIDIA | 32 GB | Runs great | Q8_0 | 11.28 GB | 140.5 t/s |
| RTX PRO 6000 Blackwell NVIDIA | 96 GB | Runs great | Q8_0 | 11.28 GB | 140.5 t/s |
| GeForce RTX 4090 NVIDIA | 24 GB | Runs great | Q8_0 | 11.28 GB | 79.1 t/s |
| GeForce RTX 3090 Ti NVIDIA | 24 GB | Runs great | Q8_0 | 11.28 GB | 79.1 t/s |
| GeForce RTX 5080 NVIDIA | 16 GB | Runs great | Q8_0 | 11.28 GB | 75.3 t/s |
| RTX 6000 Ada Generation NVIDIA | 48 GB | Runs great | Q8_0 | 11.28 GB | 75.3 t/s |
| GeForce RTX 3090 NVIDIA | 24 GB | Runs great | Q8_0 | 11.28 GB | 73.4 t/s |
| GeForce RTX 5070 Ti NVIDIA | 16 GB | Runs great | Q8_0 | 11.28 GB | 70.3 t/s |
| GeForce RTX 5090 Laptop NVIDIA | 24 GB | Runs great | Q8_0 | 11.28 GB | 70.3 t/s |
| Radeon RX 7900 XTX AMD | 24 GB | Runs great | Q8_0 | 11.28 GB | 66.1 t/s |
| Apple M3 Ultra 96GB Apple | 96 GB | Runs great | Q8_0 | 10.98 GB | 61.1 t/s |
| Apple M3 Ultra 256GB Apple | 256 GB | Runs great | Q8_0 | 10.98 GB | 61.1 t/s |
| Apple M3 Ultra 512GB Apple | 512 GB | Runs great | Q8_0 | 10.98 GB | 61.1 t/s |
| GeForce RTX 5080 Laptop NVIDIA | 16 GB | Runs great | Q8_0 | 11.28 GB | 60.2 t/s |
| RTX A6000 NVIDIA | 48 GB | Runs great | Q8_0 | 11.28 GB | 60.2 t/s |
| RTX A5000 NVIDIA | 24 GB | Runs great | Q8_0 | 11.28 GB | 60.2 t/s |
| Apple M1 Ultra 64GB Apple | 64 GB | Runs great | Q8_0 | 10.98 GB | 59.7 t/s |
| Apple M1 Ultra 128GB Apple | 128 GB | Runs great | Q8_0 | 10.98 GB | 59.7 t/s |
| Apple M2 Ultra 64GB Apple | 64 GB | Runs great | Q8_0 | 10.98 GB | 59.7 t/s |
| Apple M2 Ultra 128GB Apple | 128 GB | Runs great | Q8_0 | 10.98 GB | 59.7 t/s |
| Apple M2 Ultra 192GB Apple | 192 GB | Runs great | Q8_0 | 10.98 GB | 59.7 t/s |
| Radeon PRO W7900 AMD | 48 GB | Runs great | Q8_0 | 11.28 GB | 59.5 t/s |
| GeForce RTX 4080 SUPER NVIDIA | 16 GB | Runs great | Q8_0 | 11.28 GB | 57.7 t/s |
| GeForce RTX 4080 NVIDIA | 16 GB | Runs great | Q8_0 | 11.28 GB | 56.2 t/s |
| Radeon RX 7900 XT AMD | 20 GB | Runs great | Q8_0 | 11.28 GB | 55.1 t/s |
| GeForce RTX 4070 Ti SUPER NVIDIA | 16 GB | Runs great | Q8_0 | 11.28 GB | 52.7 t/s |
| GeForce RTX 4090 Laptop NVIDIA | 16 GB | Runs great | Q8_0 | 11.28 GB | 45.2 t/s |
| Radeon RX 9070 XT AMD | 16 GB | Runs great | Q8_0 | 11.28 GB | 44.4 t/s |
| Radeon RX 9070 AMD | 16 GB | Runs great | Q8_0 | 11.28 GB | 44.4 t/s |
| Radeon RX 7800 XT AMD | 16 GB | Runs great | Q8_0 | 11.28 GB | 43 t/s |
| Apple M4 Max 36GB Apple | 36 GB | Runs great | Q8_0 | 10.98 GB | 40.7 t/s |
| Apple M4 Max 48GB Apple | 48 GB | Runs great | Q8_0 | 10.98 GB | 40.7 t/s |
| Apple M4 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 10.98 GB | 40.7 t/s |
| Apple M4 Max 128GB Apple | 128 GB | Runs great | Q8_0 | 10.98 GB | 40.7 t/s |
| Radeon RX 7900 GRE AMD | 16 GB | Runs great | Q8_0 | 11.28 GB | 39.7 t/s |
| Radeon RX 6900 XT AMD | 16 GB | Runs great | Q8_0 | 11.28 GB | 35.3 t/s |
| Radeon RX 6800 AMD | 16 GB | Runs great | Q8_0 | 11.28 GB | 35.3 t/s |
| GeForce RTX 5060 Ti 16GB NVIDIA | 16 GB | Runs great | Q8_0 | 11.28 GB | 35.1 t/s |
| RTX A4000 NVIDIA | 16 GB | Runs great | Q8_0 | 11.28 GB | 35.1 t/s |
| Arc A770 16GB Intel | 16 GB | Runs great | Q8_0 | 11.28 GB | 34.8 t/s |
| Apple M1 Max 32GB Apple | 32 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| Apple M1 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| Apple M2 Max 32GB Apple | 32 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| Apple M2 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| Apple M2 Max 96GB Apple | 96 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| Apple M3 Max 36GB Apple | 36 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| Apple M3 Max 48GB Apple | 48 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| Apple M3 Max 64GB Apple | 64 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| Apple M3 Max 96GB Apple | 96 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| Apple M3 Max 128GB Apple | 128 GB | Runs great | Q8_0 | 10.98 GB | 29.8 t/s |
| GeForce RTX 4060 Ti 16GB NVIDIA | 16 GB | Runs well | Q8_0 | 11.28 GB | 22.6 t/s |
| NVIDIA DGX Spark 128GB NVIDIA | 128 GB | Runs well | Q8_0 | 11.28 GB | 21.4 t/s |
| Apple M4 Pro 24GB Apple | 24 GB | Runs well | Q8_0 | 10.98 GB | 20.4 t/s |
| Apple M4 Pro 48GB Apple | 48 GB | Runs well | Q8_0 | 10.98 GB | 20.4 t/s |
| Apple M4 Pro 64GB Apple | 64 GB | Runs well | Q8_0 | 10.98 GB | 20.4 t/s |
| Radeon RX 7600 XT AMD | 16 GB | Runs well | Q8_0 | 11.28 GB | 19.8 t/s |
| GeForce RTX 3080 Ti NVIDIA | 12 GB | Runs great | Q6_K | 9.19 GB | 89.4 t/s |
| GeForce RTX 3080 12GB NVIDIA | 12 GB | Runs great | Q6_K | 9.19 GB | 89.4 t/s |
| GeForce RTX 3080 10GB NVIDIA | 10 GB | Just fits | Q6_K | 9.19 GB | 74.5 t/s |
| GeForce RTX 5070 NVIDIA | 12 GB | Runs great | Q6_K | 9.19 GB | 65.8 t/s |
What to buy to run Gemma 2 9B Instruct
The cheapest hardware that runs it at a quantisation worth using and a speed you would not resent, at an 8k context window.
Cheapest that works
GeForce RTX 2060 12GB
$160 used
Q6_K · uses 9.19 GB of its 12 GB · about 32.9 tokens/s
Best value
GeForce RTX 3050 8GB
$170 used
IQ4_XS · about 31.2 tokens/s · 224 GB/s
| Device | Price | Memory | Runs it at | Speed | |
|---|---|---|---|---|---|
| GeForce GTX 1080 Ti NVIDIA | $150 used | 11 GB | Q6_K | 47.4 t/s | Check price |
| GeForce RTX 2060 12GB NVIDIA | $160 used | 12 GB | Q6_K | 32.9 t/s | Check price |
| Arc A750 Intel | $200 used | 8 GB | IQ4_XS | 56.6 t/s | Check price |
| GeForce RTX 3060 Ti NVIDIA | $230 used | 8 GB | IQ4_XS | 62.4 t/s | Check price |
| Arc A770 16GB Intel | $280 used | 16 GB | Q8_0 | 34.8 t/s | Check price |
| GeForce RTX 3070 Ti NVIDIA | $300 used | 8 GB | IQ4_XS | 84.7 t/s | Check price |
| GeForce RTX 3080 12GB NVIDIA | $400 used | 12 GB | Q6_K | 89.4 t/s | Check price |
| Radeon RX 7900 XT AMD | $650 | 20 GB | Q8_0 | 55.1 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 catalogue
Source: google/gemma-2-9b-it on Hugging Face. Downloaded 694k times in the last month. Published 2024-06-24. Architecture figures are read from the repository's own configuration file, so they move when the model does.
Gemma 4 E4B Instruct
8B · 128k context
Gemma 4 12B Instruct
11.96B · 256k context
Gemma 3 12B Instruct
12.19B · 128k context
vLLM Translategemma 12B Instruct
13.19B · 128k context
Gemma 4 E2B Instruct
5.12B · 128k context
Gemma 3 4B Instruct
4.3B · 128k context
Gemma 2 2B Instruct
2.61B · 8k context
Gemma 3 1B Instruct
1B · 32k context