Gemma 4 31B: VRAM requirements & speed
Gemma 4 31B (31.27B dense, apache-2.0). Native context 256k. Smallest GGUF: 18.32 GB.
Memory needed by quantization and context
GGUF file size plus f16 KV cache (and fixed overhead of ~1 GB per GPU not shown). Source: unsloth/gemma-4-31B-it-GGUF.
| Quant | File | +8k ctx | +32k ctx | +128k ctx | +256k ctx |
|---|---|---|---|---|---|
| Q4_K_M | 18.32 GB | 19.8 GB | 21.8 GB | 29.9 GB | 40.6 GB |
| Q5_K_M | 21.66 GB | 23.2 GB | 25.2 GB | 33.2 GB | 44.0 GB |
| Q6_K | 25.2 GB | 26.7 GB | 28.7 GB | 36.8 GB | 47.5 GB |
| Q8_0 | 33.15 GB | 34.7 GB | 36.7 GB | 44.7 GB | 55.5 GB |
KV cache size by context length
Computed from the model's attention layout (sliding-window layers capped at their window).
| Context | f16 | q8_0 | q4_0 |
|---|---|---|---|
| 8k | 1.51 GB | 0.80 GB | 0.42 GB |
| 32k | 3.52 GB | 1.87 GB | 0.99 GB |
| 128k | 11.58 GB | 6.15 GB | 3.26 GB |
| 256k | 22.31 GB | 11.85 GB | 6.28 GB |
Which hardware runs Gemma 4 31B? · 16k context
| Hardware | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|
| RTX 5090 32GB | Q6_K | fits fully | 27.4 GB | 55 | measured |
| RTX 4090 24GB | Q4_K_M | fits fully | 20.5 GB | 44 | measured |
| AMD RX 7900 XTX 24GB | Q4_K_M | fits fully | 20.5 GB | 40 | scaled from measured |
| RTX 3090 24GB | Q4_K_M | fits fully | 20.5 GB | 39 | measured |
| RTX PRO 6000 96GB (bonus page) | Q8_0 | fits fully | 35.3 GB | 33 | estimated |
| Mac Studio M5 Ultra 256GB | Q8_0 | fits fully | 35.3 GB | 23 | estimated |
| Mac Studio M5 Ultra 512GB (ships late Oct 2026) | Q8_0 | fits fully | 35.3 GB | 23 | estimated |
| Mac Studio M5 Max 128GB | Q6_K | fits fully | 27.4 GB | 20 | scaled from measured |
| MacBook Pro 14/16 M5 Max 128GB | Q6_K | fits fully | 27.4 GB | 20 | scaled from measured |
| 2x RTX 3090 (48GB total) | Q8_0 | fits fully | 35.3 GB | 19 | estimated |
| Mac Studio (2025) M3 Ultra 96GB | Q4_K_M | fits fully | 20.5 GB | 18 | estimated |
| MacBook Pro 14/16 (2024) M4 Max 128GB | Q4_K_M | fits fully | 20.5 GB | 14 | estimated |
| NVIDIA DGX Spark 128GB | Q4_K_M | fits fully | 20.5 GB | 12 | estimated |
| MacBook Pro 14/16 M5 Pro 48GB | Q4_K_M | fits fully | 20.5 GB | 12 | estimated |
| AMD Ryzen AI Max+ 395 (Strix Halo) 128GB | Q4_K_M | fits fully | 20.5 GB | 10 | measured |
| RTX 5080 16GB | Q4_K_M | partial offload (slow) | 20.5 GB | 9 | estimated |
| Mac mini (2024) M4 Pro 64GB | Q4_K_M | fits fully | 20.5 GB | 9 | estimated |
| RTX 5070 Ti 16GB | Q4_K_M | partial offload (slow) | 20.5 GB | 9 | estimated |
| 16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM) | Q4_K_M | partial offload (slow) | 20.5 GB | 7 | estimated |
| Mac mini M6 32GB | Q4_K_M | fits fully | 20.5 GB | 6 | estimated |
| RTX 5070 12GB | Q4_K_M | partial offload (slow) | 20.5 GB | 6 | estimated |
| RTX 4070 12GB | Q4_K_M | partial offload (slow) | 20.5 GB | 6 | estimated |
| MacBook Air M5 32GB | Q4_K_M | fits fully | 20.5 GB | 5 | estimated |
| 8GB VRAM class (RTX 4060 8GB, 32GB RAM) | Q4_K_M | partial offload (slow) | 20.5 GB | 4 | estimated |
| RTX 3060 12GB (32GB RAM) | Q4_K_M | partial offload (slow) | 20.5 GB | 3 | estimated |
| Mac mini / iMac M4 16GB | - | does not fit | - | - |
Discrete GPUs are modelled with 32 GB of system RAM available for offload. Does not fit even with offload: Mac mini / iMac M4 16GB.
Links
Model card · GGUF files · Open in the advisor · last verified 2026-10-14
FAQ
How much VRAM does Gemma 4 31B need?
The smallest quant (Q4_K_M) is 18.32 GB, so you need roughly 21 GB for an 8k context including KV cache and overhead. At 256k context the f16 KV cache alone is 22.3 GB.
Can I run Gemma 4 31B on a RTX 5090 32GB?
Yes: on RTX 5090 32GB it fits fully as Q6_K at about 55 tok/s (measured). 18 of 26 hardware configurations fit it fully at 16k context.
What is the maximum context for Gemma 4 31B?
256k tokens natively. Memory, not the model, is usually the limit: see the KV cache table above.
All models: model index · by memory: 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB. Numbers use the same fit math as the advisor; sources and methods.