Qwen3.6-35B-A3B-MTP requirements — can you run it?
~35.5B parameters · 21 quantizations measured · updated 2026-05-20 · source: unsloth/Qwen3.6-35B-A3B-MTP-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_M | 11.4 GB | 15.1 GB | RTX 3090 24GB |
| UD-IQ2_XXS | 11.8 GB | 15.7 GB | RTX 3090 24GB |
| UD-IQ2_M | 11.9 GB | 15.8 GB | RTX 3090 24GB |
| UD-Q2_K_XL | 12.6 GB | 16.6 GB | RTX 3090 24GB |
| UD-IQ3_XXS | 14.1 GB | 18.4 GB | RTX 3090 24GB |
| UD-IQ3_S | 15.3 GB | 19.9 GB | RTX 3090 24GB |
| UD-Q3_K_M | 17.1 GB | 22.0 GB | RTX 3090 24GB |
| UD-Q3_K_XL | 17.2 GB | 22.2 GB | RTX 5090 32GB |
| UD-IQ4_XS | 18.2 GB | 23.4 GB | RTX 5090 32GB |
| UD-IQ4_NL | 18.5 GB | 23.7 GB | RTX 5090 32GB |
| UD-Q4_K_S | 21.4 GB | 27.2 GB | RTX 5090 32GB |
| UD-Q4_K_M | 22.7 GB | 28.7 GB | RTX 5090 32GB |
| UD-Q4_K_XL | 22.9 GB | 28.9 GB | RTX 5090 32GB |
| UD-Q5_K_S | 25.5 GB | 32.1 GB | RTX PRO 6000 96GB |
| UD-Q5_K_M | 27.1 GB | 34.0 GB | RTX PRO 6000 96GB |
| UD-Q5_K_XL | 27.2 GB | 34.1 GB | RTX PRO 6000 96GB |
| UD-Q6_K | 30.0 GB | 37.5 GB | RTX PRO 6000 96GB |
| UD-Q6_K_XL | 32.6 GB | 40.6 GB | RTX PRO 6000 96GB |
| Q8_0 | 37.8 GB | 46.9 GB | RTX PRO 6000 96GB |
| UD-Q8_K_XL | 39.1 GB | 48.4 GB | RTX PRO 6000 96GB |
| BF16 | 71.1 GB | 86.8 GB | RTX PRO 6000 96GB |
KV cache uses a flat +20% context/overhead rule because this repo does not publish its config; exact numbers may differ for long contexts. MoE models keep all experts in memory — total size counts, not just active params.
Can I run Qwen3.6-35B-A3B-MTP on my GPU? (UD-IQ2_XXS recommended quant)
| Hardware | Memory | Verdict |
|---|---|---|
| RTX 3060 12GB | 12 GB | no |
| RTX 4060 Ti 16GB | 16 GB | tight |
| 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 | tight |
| 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 Qwen3.6-35B-A3B-MTP need?
The honest answer: 15.7 GB at the UD-IQ2_XXS quant with 8K context — that is the measured file size (11.8 GB) plus KV cache and runtime overhead. The smallest published build needs 15.1 GB; the lossless one needs 86.8 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 Qwen3.6-35B-A3B-MTP need?
The measured answer: 15.7 GB at the UD-IQ2_XXS quant with 8K context. The smallest published build needs 15.1 GB; the lossless (F16/BF16) build needs 86.8 GB. These are real GGUF file sizes from Hugging Face, not formula estimates.
Can I run Qwen3.6-35B-A3B-MTP on a 24GB GPU (RTX 3090/4090)?
Yes — at UD-IQ2_XXS (15.7 GB). Quants that fit 24GB: UD-IQ1_M, UD-IQ2_XXS, UD-IQ2_M, UD-Q2_K_XL, UD-IQ3_XXS, UD-IQ3_S, UD-Q3_K_M, UD-Q3_K_XL, UD-IQ4_XS, UD-IQ4_NL.
Can I run Qwen3.6-35B-A3B-MTP on CPU with system RAM?
Yes, if your usable RAM is at least the file size. At UD-IQ2_XXS you need ~15.7 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 Qwen3.6-35B-A3B-MTP 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.