CanItRun Logocanitrun.

Gemma 4 E4B

Gemma 4 E4B needs roughly 5.3 GB VRAM at Q4_K_M quantization (17.0 GB at FP16). 117 GPUs we track can run it fully in VRAM at 8k context.

117 GPUs run this natively · 1 with CPU offload

Google7.52B params128k contextApache 2.0Commercial use ok

Gemma 4 E4B is a 7.52B parameter dense model developed by Google. Released 2 April 2026 under Apache 2.0 as one of Google's two on-device Gemma 4 models. The E stands for effective parameters: 4.5B run per token, but the checkpoint stores about 8B once its Per-Layer Embedding (PLE) tables are counted (the text-only GGUF holds 7.52B), and those tables must sit in memory even though each token only looks up a few rows. 42 layers in a repeating pattern of five 512-token sliding-window layers and one full-attention layer, with the last 18 layers sharing earlier layers' KV cache. Text, image, and audio input at 128K context.

To run Gemma 4 E4B locally: Size the card for the 7.52B of loaded weights, not the 4.5B in the name: Q4_K_M weights are about 4.6 GB and the total is 5.30 GB at 8k context, Q5_K_M is 6.17 GB, and Q6_K is 7.09 GB, so an 8 GB card tops out at Q6_K. Published GGUF files agree: Q4_K_M is 4.98 GB from unsloth and 5.41 GB from bartowski, and Q8_0 is about 8.0 GB. Because sliding-window layers and KV sharing leave only 24 layers with a cache, the full 131,072-token window adds just 2.17 GB of KV cache. Decode speed follows the 4.5B effective parameters, so it runs faster than a dense 7.5B model would. Vision and audio load from a separate mmproj file; Ollama's gemma4:e4b download is 6.6 GB.

Google reports MMLU-Pro 69.4, GPQA Diamond 58.6, and LiveCodeBench v6 52.0, a large step over the Gemma 3 4B it replaces. These are vendor-measured figures from the model card.

VRAM at each quantization

Figures below assume 8k context. This model's hybrid attention stack caches only some layers, so KV cache grows much slower than linearly as context increases.

QuantWeightsKV cacheTotal
FP3230.1 GB0.16 GB33.9 GB
BF1615.0 GB0.16 GB17.0 GB
FP1615.0 GB0.16 GB17.0 GB
Q8_08.0 GB0.16 GB9.1 GB
Q6_K6.2 GB0.16 GB7.1 GB
Q5_K_Mrec5.3 GB0.16 GB6.2 GB
Q4_K_M4.6 GB0.16 GB5.3 GB
Q3_K_M3.6 GB0.16 GB4.2 GB
Q2_K2.9 GB0.16 GB3.4 GB
NVFP4cuda3.8 GB0.16 GB4.4 GB

KV cache is calculated at 8k context (FP16). Note that NVFP4 only runs on CUDA GPUs. Turn on TurboQuant in the calculator above for lower KV cache estimates.

Benchmarks

MMLU-Pro
69.4
GPQA Diamond
58.6
LiveCodeBench
52.0

GPUs that run Gemma 4 E4B natively (117)

Show 112 more
Plus 1 GPU that runs it with CPU offload (slower)

No GPU? Rent an RTX 3080 10GB from $0.17/hr ↗

Prices as of Oct 17, 2026. Rates change; check RunPod for current pricing. Affiliate link: CanItRun may earn a commission. Doesn't affect the fit calculation above.

Notes

The E stands for effective parameters: 4.5B run per token, but about 7.5B of text weights load (Per-Layer Embedding tables are resident yet only looked up), and the audio and vision encoders add roughly 0.45B more in the separate mmproj file. Ollama's Q4_K_M is a 6.6 GB download for that reason. Sliding-window attention (512 tokens) plus KV sharing across the last 18 layers keeps the long-context cache small.

Hugging Face ↗Ollama ↗Released 2026-04-02

Continue reading

vram-guides8 min
Best LLMs You Can Run on 8 GB VRAM (2026)
vram-guides8 min
Best LLMs for 6 GB VRAM (2026)

Frequently asked questions

What are the VRAM requirements for Gemma 4 E4B?
Gemma 4 E4B requires approximately 5.3 GB of VRAM at Q4_K_M quantization, 9.1 GB at Q8, and 17.0 GB at FP16. These numbers assume 8k context window; its hybrid attention stack caches far fewer than all layers, so VRAM grows much slower than linearly with context.
How many parameters does Gemma 4 E4B have?
Gemma 4 E4B has 7.52 billion parameters.
How capable is Gemma 4 E4B?
With an MMLU-Pro score of 69.4, Gemma 4 E4B delivers solid general-purpose performance suitable for most everyday tasks and professional use.
Can Gemma 4 E4B run on a 16 GB GPU?
Yes. Gemma 4 E4B needs 5.3 GB at Q4_K_M, which fits in a 16 GB GPU like the RTX 4080 or RTX 5070 Ti.
What is the smallest quantization for Gemma 4 E4B that fits in 24 GB of VRAM?
At BF16, Gemma 4 E4B needs 17.0 GB, the highest-quality quantization that fits in 24 GB of VRAM.
What GPU do I need to run Gemma 4 E4B locally?
A 16 GB GPU is enough. At Q4_K_M, Gemma 4 E4B needs 5.3 GB VRAM. Good options: RTX 4080 (16 GB), RTX 5070 Ti (16 GB).
Can I run Gemma 4 E4B in the cloud?
Yes. The cheapest RunPod setup we found that holds Gemma 4 E4B entirely in GPU memory at 8k context is 1× RTX 3080 10GB at Q8_0, at roughly 101 tokens per second and from $0.17 per hour for the whole pod. Prices as of Oct 17, 2026; RunPod rates change, so check their pricing page for the current figure.