Gemma 3 27B IT VRAM & Memory Requirements
Specific checkpoints with sourced architecture and weight files. GGUF Q4_0 (Google QAT): 17.20 GiB–17.97 GiB at 8,192 tokens, 1 sequence, FP16 KV, llama.cpp.
This result is an estimate, not a measured trace.Not verifiedFile-backed weights
Publisher and provenance
- PublisherGoogle
- Hugging Face
google/gemma-3-27b-it - Revision
005ad3404e59d6023443cb575daa05336842228a - Total parameters27,432,406,640
- Layers62
- Query heads32
- KV heads16 (sliding-window)
- Head dimension128
- Recorded context capability131,072
- Completenesscomplete · full
Memory layers in this estimate
Artifact bytes, raw KV, engine extras, and total demand are separate. This is not a runtime trace.
- Weight artifact
- 16.05 GiB (17,229,630,496 bytes)
- Raw KV cache
- 1.03 GiB (1,107,296,256 bytes)
- Engine extras / workspace
- 0.13 GiB–0.89 GiB
- Total estimated demand
- 17.20 GiB–17.97 GiB
Published artifacts
| Quantization | Selected files | Bytes | Engine | Evidence |
|---|---|---|---|---|
| Official BF16 safetensors (text + SigLIP vision tower) bf16 | 54,864,980,440 (51.10 GiB / 54.86 GB) | vllm, sglang, unspecified | file-metadata | |
| GGUF Q4_0 (Google QAT) gguf-q4_0 | gemma-3-27b-it-q4_0.gguf | 17,229,630,496 (16.05 GiB / 17.23 GB) | llama.cpp, ollama, unspecified | file-metadata |
| INT4 theoretical | No published file | 13,716,203,320 (theoretical) | — | theoretical |
Memory by quantization and context
Concurrent sequences = 1. KV defaults to FP16. The range includes documented runtime extras for the listed engine. These figures use the same core as the interactive calculator.
| Quantization | 2,048 tokens | 8,192 tokens | 32,768 tokens |
|---|---|---|---|
| Official BF16 safetensors (text + SigLIP vision tower) | 52.28 GiB–54.18 GiB | 52.75 GiB–54.66 GiB | 54.63 GiB–56.59 GiB |
| GGUF Q4_0 (Google QAT) | 16.73 GiB–17.44 GiB | 17.20 GiB–17.97 GiB | 19.08 GiB–19.98 GiB |
| INT4 theoretical Theoretical rows are not downloadable artifacts. | 13.46 GiB–17.69 GiB | 13.93 GiB–18.30 GiB | 15.81 GiB–20.60 GiB |
Raw KV contribution
Related pages
Sources
- Gemma 3 model card (Google AI for Developers) — Input context 128K tokens for 4B, 12B, 27B and 32K for 1B; images encoded to 256 tokens each. License: Gemma Terms of Use.
- Google gemma_pytorch config.py (architecture) — get_config_for_27b_v3: num_hidden_layers=62, num_attention_heads=32, num_key_value_heads=16, hidden_size=5376, head_dim=128, sliding_window_size=1024, attn_types = 5 LOCAL_SLIDING then 1 GLOBAL, cycled by layer index (gemma/model.py: attn_types[i % len(attn_types)]), so 10 global and 52 sliding layers. The Hugging Face config.json is gated (HTTP 401), so it was not read.
- Cross-check only: llama.cpp org GGUF header converted from Google's config (not an official Google source) — Header read with an HTTP range request on 2026-10-18: block_count 62, head_count 32, head_count_kv 16, embedding_length 5376, key_length 128, sliding_window 1024, context_length 131072. All agree with gemma_pytorch config.py.
- Gemma releases — March 10, 2025: release of Gemma 3 in 1B, 4B, 12B and 27B sizes.
- Hugging Face API (BF16) — sha 005ad3404e59d6023443cb575daa05336842228a; gated manual; safetensors.total=27432406640; 12 safetensors siblings.
- Hugging Face API (QAT Q4_0 GGUF) — sha 17cf0f6ad611f1a57a1640daa57eb427d6e67ed6; gemma-3-27b-it-q4_0.gguf = 17,229,630,496 bytes.
Assumptions that affect this result
- Requested context is within the recorded native window of 131072 tokens.
- Full-resident weight estimates use total parameters, including inactive MoE experts.
- Weight bytes come from the selected weight files only, not from every file in the repository.
- Download file size is not peak GPU memory. Runtime layout, allocator padding, and KV are extra.
- Counted files: gemma-3-27b-it-q4_0.gguf.
- This checkpoint has a vision component. Text-only inference does not include vision encoder or projector bytes.
- Context budget is the number of cached tokens per sequence (prompt plus reserved generation).
- Concurrent sequences share one model replica; weights are not multiplied by sequence count.
Calculator
Loading catalog snapshot…
Search or pick a catalog checkpoint to see memory demand.
Search or pick a catalog checkpoint to see memory demand.