GeForce RTX 3080 (10 GB)
NVIDIA
CompatibleRuns Gemma 2 9B IT Q4_K_M entirely in memory
Model · Google
Gemma 2 9B IT: VRAM and hardware requirements
At Q4_K_M with an 8K context, Gemma 2 9B IT needs about 8.6 GB: a GPU with 10 GB or more runs it entirely in VRAM.
- Parameters9.24B
- Max context8K
- LicenseGemma Terms of Use
Specifications
- Organization
- Architecture
- gemma2
- Released
- June 2024
- Layers
- 42
- Hidden size
- 3584
- Attention heads
- 16
- KV heads
- 8
- Head dimension
- 256
- Vocabulary
- 256,000
- KV cache per 1K tokens (f16)
- 0.328 GB
- Repository
- google/gemma-2-9b-it
Official specData updated 24 days agoSource: Hugging Face config.json + model card
Quantizations and memory
| Quantization | File size | Memory needed | Minimum GPU memory | Quality |
|---|---|---|---|---|
| BF16 | Not available | ~21.2 GB | 24 GB | Reference (unquantized) |
| Q8_0 | 9.2 GB | ~12.7 GB | 16 GB | Near-lossless |
| Q6_K | 7.1 GB | ~10.5 GB | 11 GB | Near-lossless |
| Q5_K_M | 6.2 GB | ~9.5 GB | 10 GB | Very good |
| Q4_K_MMost used | 5.4 GB | ~8.6 GB | 10 GB | Good — the usual choice |
| IQ4_XS | 4.8 GB | ~8.1 GB | 10 GB | Good — the usual choice |
| Q3_K_M | 4.4 GB | ~7.6 GB | 10 GB | Noticeable loss |
| Q2_K | 3.5 GB | ~6.7 GB | 8 GB | Significant loss |
File sizes: not available until imported from the model repository. Memory = weights (estimated from bits per weight) + KV cache + runtime overhead + safety margin. (BF16 ≈ 16 bpw, Q8_0 ≈ 8.5 bpw, Q6_K ≈ 6.57 bpw, Q5_K_M ≈ 5.7 bpw, Q4_K_M ≈ 4.9 bpw, IQ4_XS ≈ 4.46 bpw, Q3_K_M ≈ 4 bpw, Q2_K ≈ 3.17 bpw)
Q4, Q5 or Q8: how to choose a quantization
GPUs that run Gemma 2 9B IT
| RTX 509032 GB · 64 GB RAM~8.7 GB · 134–213 tok/s estimate | Excellent |
|---|---|
| RTX 409024 GB · 64 GB RAM~8.7 GB · 89–130 tok/s estimate | Excellent |
| RTX 309024 GB · 64 GB RAM~8.7 GB · 84–122 tok/s estimate | Excellent |
| RX 7900 XTX24 GB · 64 GB RAM~8.8 GB · 70–109 tok/s estimate | Excellent |
| RTX 5060 Ti 16 GB16 GB · 64 GB RAM~8.7 GB · 45–62 tok/s estimate | Excellent |
| RTX 5070 Ti16 GB · 64 GB RAM~8.7 GB · 81–117 tok/s estimate | Excellent |
| RTX 508016 GB · 64 GB RAM~8.7 GB · 85–125 tok/s estimate | Excellent |
| RTX 4060 Ti 16 GB16 GB · 64 GB RAM~8.7 GB · 30–40 tok/s estimate | Excellent |
| RTX 4070 Ti SUPER16 GB · 64 GB RAM~8.7 GB · 64–90 tok/s estimate | Excellent |
| RX 9070 XT16 GB · 64 GB RAM~8.8 GB · 52–77 tok/s estimate | Excellent |
| RTX 3060 12 GB12 GB · 64 GB RAM~8.7 GB · 37–50 tok/s estimate | Excellent |
| RTX 407012 GB · 64 GB RAM~8.7 GB · 50–69 tok/s estimate | Excellent |
| RTX 507012 GB · 64 GB RAM~8.7 GB · 64–90 tok/s estimate | Excellent |
| RTX 40608 GB · 64 GB RAM~8.7 GB · 12–22 tok/s estimate | CPU/GPU offload |
Macs that run Gemma 2 9B IT
| M3 Ultra (80-core GPU) · 512 GBMac Studio~8.5 GB · 60–111 tok/s estimate | Excellent |
|---|---|
| M4 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~8.5 GB · 44–74 tok/s estimate | Excellent |
| M3 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~8.5 GB · 34–54 tok/s estimate | Excellent |
| M4 Pro (20-core GPU) · 64 GBMac mini, MacBook Pro 14~8.5 GB · 24–37 tok/s estimate | Excellent |
| M4 Max (32-core GPU) · 36 GBMacBook Pro 14, MacBook Pro 16~8.5 GB · 34–56 tok/s estimate | Excellent |
| M4 · 32 GBMacBook Air, MacBook Pro 14~8.5 GB · 11–17 tok/s estimate | Excellent |
| M5 · 32 GBMacBook Pro 14~8.5 GB · 14–21 tok/s estimate | Excellent |
| M3 · 24 GBMacBook Air, MacBook Pro 14~8.5 GB · 9.5–14 tok/s estimate | Compatible |
EstimateData updated 24 days agoSource: Run LLM Locally engine
Limits in plain words
The KV cache grows with context: at 8K it adds about 2 GB on top of the weights.
With under 8 GB of VRAM, use a smaller quantization or a smaller model of the same family.
Where to buy
Prices and links follow the market you choose.
Radeon RX 7600 XT
AMD
ExcellentRuns Gemma 2 9B IT Q4_K_M entirely in memory
AMD Ryzen AI Max+ 395
Unified memory PC
ExcellentRuns Gemma 2 9B IT Q4_K_M entirely in memory
No current offer imported for this market: the link opens a marketplace search.
Frequently asked questions
How much VRAM does Gemma 2 9B IT need?
About 8.6 GB at Q4_K_M with an 8K context, including the KV cache and runtime overhead. At Q8_0 it needs about 12.7 GB. A longer context adds more: about 2 GB of KV cache at 32K.
Can Gemma 2 9B IT run on a Mac?
Yes on a Mac whose GPU can use at least 8.5 GB, e.g. M3 24 GB. macOS lets the GPU use about two thirds to three quarters of unified memory by default.
How fast is Gemma 2 9B IT?
On a RTX 4090, we estimate 89–130 tok/s at Q4_K_M. This is an estimate from memory bandwidth (confidence low) until benchmarks are imported.