Can I run Gemma 4 E4B locally?
Gemma 4 E4B has 8.00 billion parameters (Hugging Face). At the popular Q4_K_M quantization it needs roughly 5.8 GB of GPU memory ESTIMATE. It has been on Hugging Face since 2026-03-02.
Memory needed by quantization
| Quant | Weights | Total needed (est.) |
|---|---|---|
| IQ1_S | 1.6 GB | 2.3 GB |
| IQ2_XXS | 2.1 GB | 2.9 GB |
| IQ2_M | 2.7 GB | 3.5 GB |
| Q2_K | 3.2 GB | 4.0 GB |
| IQ3_XXS | 3.1 GB | 3.9 GB |
| Q3_K_S | 3.6 GB | 4.5 GB |
| IQ3_M | 3.7 GB | 4.5 GB |
| Q3_K_M | 4.0 GB | 4.9 GB |
| IQ4_XS | 4.2 GB | 5.2 GB |
| Q4_0 | 4.5 GB | 5.4 GB |
| Q4_K_M | 4.8 GB | 5.8 GB |
| MXFP4 | 4.6 GB | 5.6 GB |
| Q5_K_M | 5.7 GB | 6.7 GB |
| Q6_K | 6.6 GB | 7.6 GB |
| Q8_0 | 8.5 GB | 9.6 GB |
| F16 | 16.0 GB | 17.5 GB |
KV cache by context length ESTIMATE
| Context | f16 | q8_0 | q4_0 |
|---|---|---|---|
| 4k | 0.10 GB | 0.05 GB | 0.03 GB |
| 8k | 0.15 GB | 0.08 GB | 0.04 GB |
| 16k | 0.27 GB | 0.14 GB | 0.08 GB |
| 32k | 0.51 GB | 0.27 GB | 0.14 GB |
| 64k | 0.98 GB | 0.52 GB | 0.27 GB |
| 128k | 1.92 GB | 1.02 GB | 0.54 GB |
From config.json: 7 KV-caching attention layers, 2 KV heads × 256 head dim, plus 35 sliding-window layers (512 tokens). Max context 128k. The tables above assume 4k context.
Which GPUs and Macs can run it?
✅ fits · ⚠️ fits with little headroom · ❌ doesn't fit on this one device (multiple GPUs or RAM offload may work; check your setup). Apple Silicon assumes macOS lets the GPU use ~67–75% of unified memory. All fit verdicts are estimates.
Get the weekly "new models & what it takes to run them" email
Free. One email a week. Unsubscribe anytime.