Gemma 4 E4B: VRAM requirements & speed
Gemma 4 E4B (8B dense, apache-2.0). Native context 128k. Smallest GGUF: 4.98 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-E4B-it-GGUF.
| Quant | File | +8k ctx | +32k ctx | +128k ctx |
|---|---|---|---|---|
| Q4_K_M | 4.98 GB | 5.1 GB | 5.5 GB | 7.1 GB |
| Q5_K_M | 5.48 GB | 5.6 GB | 6.0 GB | 7.6 GB |
| Q6_K | 7.07 GB | 7.2 GB | 7.6 GB | 9.2 GB |
| Q8_0 | 8.19 GB | 8.3 GB | 8.7 GB | 10.4 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 | 0.16 GB | 0.08 GB | 0.04 GB |
| 32k | 0.56 GB | 0.30 GB | 0.16 GB |
| 128k | 2.17 GB | 1.15 GB | 0.61 GB |
Which hardware runs Gemma 4 E4B? · 16k context
| Hardware | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|
| RTX 5090 32GB | Q8_0 | fits fully | 8.5 GB | 177 | scaled from measured |
| RTX PRO 6000 96GB | Q8_0 | fits fully | 8.5 GB | 174 | scaled from measured |
| RTX 5080 16GB | Q8_0 | fits fully | 8.5 GB | 105 | scaled from measured |
| RTX 5070 Ti 16GB | Q8_0 | fits fully | 8.5 GB | 99 | scaled from measured |
| RTX 3090 24GB | Q8_0 | fits fully | 8.5 GB | 97 | scaled from measured |
| RTX 4090 24GB | Q8_0 | fits fully | 8.5 GB | 96 | scaled from measured |
| Mac Studio M5 Ultra 256GB | Q8_0 | fits fully | 8.5 GB | 91 | estimated |
| Mac Studio M5 Ultra 512GB (ships late Oct 2026) | Q8_0 | fits fully | 8.5 GB | 91 | estimated |
| 2x RTX 3090 (48GB total) | Q8_0 | fits fully | 8.5 GB | 78 | estimated |
| RTX 5070 12GB | Q8_0 | fits fully | 8.5 GB | 77 | scaled from measured |
| RTX 4070 12GB | Q8_0 | fits fully | 8.5 GB | 58 | scaled from measured |
| RTX 5060 Ti 16GB | Q8_0 | fits fully | 8.5 GB | 52 | scaled from measured |
| 8GB VRAM class (RTX 4060 8GB, 32GB RAM) | Q5_K_M | fits fully | 5.8 GB | 47 | scaled from measured |
| Mac Studio M5 Max 128GB | Q8_0 | fits fully | 8.5 GB | 46 | scaled from measured |
| MacBook Pro 14/16 M5 Max 128GB | Q8_0 | fits fully | 8.5 GB | 46 | scaled from measured |
| RTX 3060 12GB (32GB RAM) | Q8_0 | fits fully | 8.5 GB | 42 | scaled from measured |
| Mac Studio (2025) M3 Ultra 96GB | Q8_0 | fits fully | 8.5 GB | 39 | scaled from measured |
| 16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM) | Q8_0 | fits fully | 8.5 GB | 35 | scaled from measured |
| MacBook Pro 14/16 (2024) M4 Max 128GB | Q8_0 | fits fully | 8.5 GB | 32 | scaled from measured |
| AMD RX 7900 XTX 24GB | Q8_0 | fits fully | 8.5 GB | 29 | scaled from measured |
| NVIDIA DGX Spark 128GB | Q8_0 | fits fully | 8.5 GB | 29 | scaled from measured |
| AMD Ryzen AI Max+ PRO 495 (Ryzen AI Halo) 192GB | Q8_0 | fits fully | 8.5 GB | 27 | estimated |
| MacBook Pro 14/16 M5 Pro 48GB | Q8_0 | fits fully | 8.5 GB | 26 | scaled from measured |
| AMD Ryzen AI Max+ 395 (Strix Halo) 128GB | Q5_K_M | fits fully | 5.8 GB | 19 | scaled from measured |
| Mac mini (2024) M4 Pro 64GB | Q8_0 | fits fully | 8.5 GB | 19 | scaled from measured |
| MacBook Air M5 32GB | Q5_K_M | fits fully | 5.8 GB | 18 | scaled from measured |
| Mac mini M6 32GB | Q6_K | fits fully | 7.4 GB | 17 | scaled from measured |
| Mac mini / iMac M4 16GB | Q4_K_M | fits fully | 5.3 GB | 16 | scaled from measured |
Discrete GPUs are modelled with 32 GB of system RAM available for offload. Fits (at least with offload) on every listed configuration.
MLX conversions for Apple Silicon
| Repo | Bits | Size | Runs with |
|---|---|---|---|
| mlx-community/gemma-4-e4b-it-4bit | 4-bit (4bit) | 5.1 GB | mlx-vlm |
| mlx-community/gemma-4-e4b-it-8bit | 8-bit (8bit) | 8.9 GB | mlx-vlm |
Verified via the Hugging Face API on 2026-10-20. Compare with the GGUF sizes above (Q4_K_M 4.98 GB, Q8_0 8.19 GB). Mac hardware pages show the MLX command; there are no MLX speeds on this site yet. MLX vs GGUF.
Links
Model card · GGUF files · Open in the advisor · last verified 2026-10-20
FAQ
How much VRAM does Gemma 4 E4B need?
The smallest quant (Q4_K_M) is 4.98 GB, so you need roughly 6 GB for an 8k context including KV cache and overhead. At 128k context the f16 KV cache alone is 2.2 GB.
Can I run Gemma 4 E4B on a RTX 5090 32GB?
Yes: on RTX 5090 32GB it fits fully as Q8_0 at about 177 tok/s (scaled from measured). 28 of 28 hardware configurations fit it fully at 16k context.
What is the maximum context for Gemma 4 E4B?
128k 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.