Gemma 3 27B VRAM Requirements

Gemma 3 27B needs about 17.6 GB of VRAM at Q4_K_M with an 8K context: 15.5 GB for weights, 1.0 GB for the KV cache and 1.1 GB of runtime overhead. At Q8_0 it needs 29.2 GB and at full FP16 precision 53.1 GB. The smallest GPU that runs it fully is the Radeon RX 7900 XT 20GB.

By Google · Updated October 14, 2026 · Values from config.json on Hugging Face

17.6 GB VRAM

Gemma 3 27B at Q4_K_M with 8K context and the KV cache in FP16.

  • Weights 15.5 GB
  • KV cache 1.0 GB
  • Runtime overhead 1.1 GB

Which GPUs can run it

Gemma 3 27B VRAM by quantization

Total memory with an FP16 KV cache. "Smallest tier" is the smallest standard VRAM size with a 6% safety margin at 8K.

QuantizationBits/weightWeightsTotal at 8KTotal at 32KSmallest tier
Q4_K_M4.8515.5 GB17.6 GB19.5 GB20 GB
Q5_K_M5.6918.1 GB20.3 GB22.1 GB24 GB
Q6_K6.5921.0 GB23.1 GB25.0 GB32 GB
Q8_08.5027.1 GB29.2 GB31.1 GB32 GB
FP16 / BF1616.0051.0 GB53.1 GB55.0 GB64 GB
Q3_K_M3.9112.5 GB14.6 GB16.5 GB16 GB
Q2_K3.009.6 GB11.7 GB13.6 GB16 GB

VRAM by context length

At Q4_K_M. 10 of 62 layers keep the full context. The other 52 use a 1024-token sliding window, so their cache stops growing after 1K tokens.

ContextKV cache (FP16)KV cache (Q8_0)Total, FP16 cacheTotal, Q8_0 cache
4K tokens0.7 GB0.4 GB17.3 GB16.9 GB
8K tokens1.0 GB0.5 GB17.6 GB17.1 GB
32K tokens2.9 GB1.5 GB19.5 GB18.1 GB
128K tokens10.4 GB5.5 GB27.0 GB22.1 GB

GPUs that can run Gemma 3 27B

At Q4_K_M with 8K context. "Best quant" is the highest-quality format that fits fully; "max context" is at Q4_K_M.

GPUUsable memoryStatusBest quantMax contextSpeed
Radeon RX 7900 XT 20GB20.0 GBFitsQ4_K_M23K~25 tok/s
Apple M5 32GB21.3 GBFitsQ4_K_M39K~5.7 tok/s
NVIDIA L4 24GB24.0 GBFitsQ5_K_M71K~11 tok/s
Tesla P40 24GB24.0 GBFitsQ5_K_M71K~11 tok/s
Apple M4 Max 36GB24.0 GBFitsQ5_K_M71K~15 tok/s
Intel Arc Pro B60 24GB24.0 GBFitsQ5_K_M71K~13 tok/s
RTX 3090 24GB24.0 GBFitsQ5_K_M71K~34 tok/s
Radeon RX 7900 XTX 24GB24.0 GBFitsQ5_K_M71K~30 tok/s
RTX 3090 Ti 24GB24.0 GBFitsQ5_K_M71K~37 tok/s
RTX 4090 24GB24.0 GBFitsQ5_K_M71K~37 tok/s
Radeon AI PRO R9700 32GB32.0 GBFitsQ8_0128K~20 tok/s
RTX 5090 32GB32.0 GBFitsQ8_0128K~64 tok/s
Apple M4 Pro 48GB36.0 GBFitsQ8_0128K~10 tok/s
A100 40GB40.0 GBFitsQ8_0128K~56 tok/s
RTX 4060 Ti 16GB16.0 GB86% on GPUQ3_K_M–~7.1 tok/s
RTX 4070 Ti Super 16GB16.0 GB86% on GPUQ3_K_M–~11 tok/s
RTX 4080 16GB16.0 GB86% on GPUQ3_K_M–~11 tok/s
RTX 4080 Super 16GB16.0 GB86% on GPUQ3_K_M–~11 tok/s
RTX 5060 Ti 16GB16.0 GB86% on GPUQ3_K_M–~9.0 tok/s
RTX 5070 Ti 16GB16.0 GB86% on GPUQ3_K_M–~12 tok/s

About Gemma 3 27B

The largest Gemma 3. It has 16 KV heads, twice as many as most models its size, but sliding windows on 52 of 62 layers offset that. Google's QAT Q4_0 release was made to fit a 24 GB consumer GPU. Older runtimes without sliding-window support store the full cache on every layer and use far more memory.

SpecValue
Parameters27.4B
Layers62
Hidden size5376
Attention heads / KV heads32 / 16
Head dimension128
Vocabulary262,208
Max context (used here)128K tokens
Sliding window1024 tokens on 52 of 62 layers
KV cache per token (FP16)80 KB
Released2025-03
LicenseGemma Terms of Use

Gemma 3 27B VRAM: common questions

Does Gemma 3 27B fit on an RTX 3090 or 4090?

Yes. At Q4_K_M it needs about 18 GB with an 8K context, and the sliding window keeps 32K context under 24 GB in runtimes that support it.

How much VRAM does Gemma 3 27B need?

About 17.6 GB at Q4_K_M with an 8K context, 29.2 GB at Q8_0 and 53.1 GB at FP16. Longer contexts add 0.08 GB per 1,000 tokens with an FP16 cache.

What is the cheapest GPU that can run Gemma 3 27B?

The smallest listed device that fits it fully at Q4_K_M with 8K context is the Radeon RX 7900 XT 20GB (20.0 GB usable).

Related