Computed with the open FitLLM engine — accurate per-layer KV-cache modeling, not a naive estimate. Updated 2026-06.
| Model weights | 15.5 GB |
| KV cache | 2.0 GB |
| Runtime overhead + reserve | 5.1 GB |
| Total used | 22.6 / 24 GB |
| Free | 1.4 GB |
Max context that fits at Q4_K_M: ~34K tokens · with Q8 KV cache → ~53K tokens.
| Weight quant | Weights | Fits (KV F16) | Used @32K |
|---|---|---|---|
| Q4_K_M | 15.5 GB | ✅ up to 34K ctx | 22.6 / 24.0 GB |
| Q5_K_M | 18.1 GB | ❌ up to 6K ctx | 25.5 / 24.0 GB |
| Q6_K | 20.8 GB | ❌ won't fit | 28.6 / 24.0 GB |
| Q8_0 | 26.9 GB | ❌ won't fit | 35.4 / 24.0 GB |
| FP16 | 50.7 GB | ❌ won't fit | 62.0 / 24.0 GB |
KV cache is F16 here (llama.cpp default). Drop it to Q8/Q4 (-ctk/-ctv) for more context.
Qwen 3.6 27B is a hybrid model: only 16 of its 64 layers use full attention — the rest are linear and keep no growing KV cache. Naive calculators count every layer at full context and badly over-estimate.
same GPU Models that fit on the RTX 3090: Qwen 3.6 27B, Gemma 4 e2b, Gemma 4 e4b, Gemma 4 12b, Gemma 4 26b A4B.
same model GPUs that run Qwen 3.6 27B: RTX 5090 (32GB), RTX 4090 (24GB), RTX 3090 (24GB), RTX 3090 Ti (24GB).
Qwen 3.6 27B = 27.2B, 64 layers. The RTX 3090 has 24GB / 936GB/s. Same math, open source: fitllm-engine. GGUF bpw from llama.cpp.
All numbers are computed by the open-source fitllm-engine (MIT) from official model config.json values — reproduce or audit them yourself. Estimates; real usage varies with runtime (llama.cpp / MLX / Ollama), driver and display. Found a mismatch? Report it. · FitLLM home