Qwen3.8-27B requirements — can you run it?
~27.3B parameters · 25 quantizations measured · updated 2026-08-20 · source: unsloth/Qwen3.8-27B-GGUF
Every quantization: real file size vs what you actually need
| Quant | File (measured) | VRAM/RAM needed (8K ctx) | Runs on (first fit) |
|---|---|---|---|
| UD-IQ1_S | 6.2 GB | 9.3 GB | RTX 3060 12GB |
| UD-IQ1_M | 6.7 GB | 9.9 GB | RTX 3060 12GB |
| UD-IQ2_XXS | 7.3 GB | 10.4 GB | RTX 3060 12GB |
| UD-IQ2_S | 8.4 GB | 11.5 GB | RTX 4060 Ti 16GB |
| UD-Q2_K_XL | 9.8 GB | 13.0 GB | RTX 4060 Ti 16GB |
| UD-IQ3_XXS | 10.9 GB | 14.1 GB | RTX 4060 Ti 16GB |
| UD-IQ3_S | 12.0 GB | 15.2 GB | RTX 3090 24GB |
| UD-Q3_K_XL | 13.1 GB | 16.3 GB | RTX 3090 24GB |
| UD-IQ4_XS | 14.3 GB | 17.5 GB | RTX 3090 24GB |
| UD-Q4_K_S | 15.4 GB | 18.7 GB | RTX 3090 24GB |
| Q4_0 | 16.1 GB | 19.5 GB | RTX 3090 24GB |
| UD-Q4_K_M | 16.5 GB | 19.9 GB | RTX 3090 24GB |
| Q4_1 | 17.5 GB | 21.1 GB | RTX 3090 24GB |
| UD-Q4_K_XL | 17.6 GB | 21.1 GB | RTX 3090 24GB |
| UD-Q5_K_S | 18.7 GB | 22.3 GB | RTX 5090 32GB |
| UD-Q5_K_M | 19.8 GB | 23.5 GB | RTX 5090 32GB |
| UD-Q5_K_XL | 20.9 GB | 24.7 GB | RTX 5090 32GB |
| UD-Q6_K | 22.0 GB | 25.9 GB | RTX 5090 32GB |
| UD-Q6_K_M | 23.1 GB | 27.1 GB | RTX 5090 32GB |
| UD-Q6_K_L | 24.2 GB | 28.3 GB | RTX 5090 32GB |
| UD-Q6_K_XL | 25.3 GB | 29.5 GB | RTX PRO 6000 96GB |
| UD-Q8_K_L | 28.0 GB | 32.4 GB | RTX PRO 6000 96GB |
| Q8_0 | 29.0 GB | 33.5 GB | RTX PRO 6000 96GB |
| UD-Q8_K_XL | 31.5 GB | 36.1 GB | RTX PRO 6000 96GB |
| BF16 | 54.7 GB | 61.2 GB | RTX PRO 6000 96GB |
KV cache computed exactly from the model config (GQA formula, 8K context). MoE models keep all experts in memory — total size counts, not just active params.
Can I run Qwen3.8-27B on my GPU? (UD-Q2_K_XL recommended quant)
| Hardware | Memory | Verdict |
|---|---|---|
| RTX 3060 12GB | 12 GB | no |
| RTX 4060 Ti 16GB | 16 GB | RUNS |
| RTX 3090 24GB | 24 GB | RUNS |
| RTX 4090 24GB | 24 GB | RUNS |
| RTX 5090 32GB | 32 GB | RUNS |
| RTX PRO 6000 96GB | 96 GB | RUNS |
| Mac 16GB unified | 16 GB | RUNS |
| Mac 32GB unified | 32 GB | RUNS |
| Mac 64GB unified | 64 GB | RUNS |
| Mac 128GB unified | 128 GB | RUNS |
| Mac 256GB unified | 256 GB | RUNS |
| Mac 512GB unified | 512 GB | RUNS |
| 32GB system RAM (CPU) | 32 GB | RUNS |
| 64GB system RAM (CPU) | 64 GB | RUNS |
| 128GB system RAM (CPU) | 128 GB | RUNS |
| 256GB system RAM (CPU) | 256 GB | RUNS |
| 1TB server RAM (CPU) | 1024 GB | RUNS |
| 2TB server RAM (CPU) | 2048 GB | RUNS |
How much VRAM does Qwen3.8-27B need?
The honest answer: 13.0 GB at the UD-Q2_K_XL quant with 8K context — that is the measured file size (9.8 GB) plus KV cache and runtime overhead. The smallest published build needs 9.3 GB; the lossless one needs 61.2 GB.
Can I run it with CPU offload?
Yes, if combined RAM+VRAM ≥ file size — but generation speed is limited by memory bandwidth. Expect roughly: DDR5 dual channel ~10-30 tok/s for small MoE actives, single digits for big ones, 0.1-1 tok/s when experts page from disk. CPU-only is fine for batch jobs, painful for chat.
Why do other sites show different numbers?
Most pages compute weights from a formula (params × bits/8). We use the actual file sizes published in GGUF repos, which include the embedding table, unquantized tensors, and container overhead — that is why our numbers can differ from a naive calculation.
FAQ
How much VRAM does Qwen3.8-27B need?
The measured answer: 13.0 GB at the UD-Q2_K_XL quant with 8K context. The smallest published build needs 9.3 GB; the lossless (F16/BF16) build needs 61.2 GB. These are real GGUF file sizes from Hugging Face, not formula estimates.
Can I run Qwen3.8-27B on a 24GB GPU (RTX 3090/4090)?
Yes — at UD-Q2_K_XL (13.0 GB). Quants that fit 24GB: UD-IQ1_S, UD-IQ1_M, UD-IQ2_XXS, UD-IQ2_S, UD-Q2_K_XL, UD-IQ3_XXS, UD-IQ3_S, UD-Q3_K_XL, UD-IQ4_XS, UD-Q4_K_S, Q4_0, UD-Q4_K_M, Q4_1, UD-Q4_K_XL, UD-Q5_K_S, UD-Q5_K_M.
Can I run Qwen3.8-27B on CPU with system RAM?
Yes, if your usable RAM is at least the file size. At UD-Q2_K_XL you need ~13.0 GB of RAM. CPU generation is memory-bandwidth-bound: expect single-digit tok/s for large MoE models, 10-30 tok/s for small active params, and below 1 tok/s when experts page from disk.
Why do VRAM numbers for Qwen3.8-27B differ between sites?
Most sites compute weights from a formula (params x bits/8). ModelFit uses the actual GGUF file sizes published on Hugging Face, which include the embedding table, unquantized tensors, and container overhead — so our numbers reflect what you really download and load.