Best home rig for Qwen3.6 27B

Will it fit my GPU? Best value GPU per euro →

Real, measured generation speeds for Qwen3.6 27B on home hardware, fastest first. Different quants/engines aren't directly comparable.

Build	Quant	Engine	tok/s
RTX 5090: 2026 models (llamabench community)	llama-bench tg128 (quant per result)	llama.cpp	75.7	source
RTX 5090 (llm-speed signed runs, Oct 2026): gpt-oss 20B 324 tok/s, Qwen3.8 27B 75 tok/s	Q4_K_M	llama.cpp 0.7.0-dev	75.1	source
Radeon AI Pro R9700: Qwen3.6 27B on Vulkan, HIP and Ollama, with and without MTP	Q4_K_M + MTP draft head (draft-n-max 3), 64K context	llama.cpp Vulkan	66.0	source
RTX 4090: 2026 models (llamabench community)	llama-bench tg128 (quant per result)	llama.cpp	47.0	source
RTX 3090 (llm-speed signed runs, Oct 2026): gpt-oss 20B 207 tok/s, Qwen3.6 27B 43 tok/s	Q4_K_M	llama.cpp 0.7.0-dev	42.7	source
RTX 3090: 2026 models (llamabench community)	llama-bench tg128 (quant per result)	llama.cpp	40.2	source
DGX Spark with Qwen3.6 27B NVFP4 + DFlash speculative decoding	NVFP4	vLLM + DFlash spec decoding	32.83	source
Radeon AI Pro R9700: Qwen3.6 27B on Vulkan, HIP and Ollama, with and without MTP	Q4_K_M, no MTP	llama.cpp Vulkan	31.9	source
Radeon AI Pro R9700: Qwen3.6 27B on Vulkan, HIP and Ollama, with and without MTP	Q4_K_M	llama.cpp HIP (ROCm)	27.1	source
Radeon AI Pro R9700: Qwen3.6 27B on Vulkan, HIP and Ollama, with and without MTP	Q4_K_M	Ollama (HIP)	26.7	source
HP ZBook Ultra G1a (Strix Halo laptop, 128 GB): Qwen3.6 35B-A3B at 59.4 tok/s, gpt-oss 120B at 44.7	4-bit	Ollama 0.34.0	11.4	source
DGX Spark with Qwen3.6 27B NVFP4 + DFlash speculative decoding	FP8	vLLM autoregressive baseline	7.85	source

Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)

Memory math: how much VRAM / RAM does Qwen3.6 27B need?

Qwen3.6 27B has 27.32B parameters (source: GGUF param count via llamabench.ai). Weight size = parameters × bits-per-weight ÷ 8, before context (KV cache) and runtime overhead.

Format	Bits/weight	Weights	Smallest common memory size that holds the weights alone
Q8_0	8.50	29.0 GB	32 GB
Q6_K	6.56	22.4 GB	24 GB
Q5_K_M	5.70	19.5 GB	24 GB
Q4_K_M	4.89	16.7 GB	24 GB
Q3_K_M	4.00	13.6 GB	16 GB
Q2_K	3.16	10.8 GB	12 GB

Bits-per-weight for GGUF quants are llama.cpp's own figures (llama.cpp quantize README). "As measured" rows apply the source's stated bits-per-weight to the parameter count; real files differ slightly because some layers stay at higher precision. Try other combinations in the VRAM calculator.

FAQ

How much VRAM does Qwen3.6 27B need?

The weights are about 16.7 GB at Q4_K_M (27.32B params × 4.89 bits ÷ 8, before context/KV cache). Add memory for context (KV cache) and the runtime. The smallest-memory setup with a measured result here, fully in memory, is RTX 4090 (24 GB) at 47.0 tok/s.

What is the fastest measured hardware for Qwen3.6 27B?

RTX 5090: 75.7 tok/s generation (llama.cpp, llama-bench tg128 (quant per result)), from https://llamabench.ai/browse/rtx5090.

Which GPUs have measured results for Qwen3.6 27B?

DGX Spark (GB10), RTX 3090, RTX 4090, RTX 5090, Radeon AI Pro R9700, Strix Halo (Radeon 8060S).

Can it run on… DGX Spark (GB10) · RTX 3090 · RTX 4090 · RTX 5090 · Radeon AI Pro R9700 · Strix Halo (Radeon 8060S)