Can I Run Qwen3.8 on Framework Desktop (Ryzen AI Max+ 395, 128 GB)?

作者： Jakub Rusinowski · 最后更新： 2026年9月11日

Yes

Yes — Qwen3.8-Flash-Next at Q4_K_M needs about 112.7 GB of the 128 GB usable on Framework Desktop (Ryzen AI Max+ 395, 128 GB) (~15.3 GB spare), at ~33.8 tok/s (estimated), with room for about 32,768 tokens of context.

Confidence: medium · Recommended quantization: Q4_K_M · Estimated speed: ~33.8 tok/s

Buy This HardwareApple Mac Studio M4 Max — 128 GB VRAM · 35 W board power Deploy in the Cloud NowRunPod

or compare on Vast.ai

As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.

Framework Desktop (Ryzen AI Max+ 395, 128 GB) — what it gives a model

Usable memory for models	128 GB
Memory bandwidth	256 GB/s
Form factor	Mini PC
Operating system	Windows or Linux
Memory upgradeable	No — soldered

Qwen3.8 on Framework Desktop (Ryzen AI Max+ 395, 128 GB): memory by quantization

Quant	Memory needed	Fits 128 GB?	Max context	Est. speed	Download
F16	364.1 GB	✗ No	—	—	360 GB
Q8_0	195.3 GB	✗ No	—	—	191.3 GB
Q6_K	151.7 GB	✗ No	—	—	147.6 GB
Q5_K_M	131.6 GB	✗ No	—	—	127.6 GB
Q4_K_M	112.7 GB	✓ Yes	32K	~33.8 tok/s	108.7 GB
Q3_K_M	80.8 GB	✓ Yes	64K	~41.6 tok/s	76.7 GB
Q2_K	63.2 GB	✓ Yes	128K	~47.6 tok/s	59.2 GB

Which Qwen3.8 sizes fit

Variant	Needs at Q4_K_M	Fits?	Est. speed
Qwen3.8-Max	1457.5 GB	✗ Too large	—
Qwen3.8-Flash-Next	112.7 GB	✓ Fits	~33.8 tok/s
Qwen3.8 27B	19.3 GB	✓ Fits	~10.6 tok/s

What to watch out for

1 larger variant of Qwen3.8 does not fit and would need CPU offload or different hardware.
This model does not publish its full attention configuration, so the KV-cache share of these figures is inferred from its parameter count rather than computed exactly.
Memory on this machine is not upgradeable, so the configuration you buy is the ceiling for every model you will ever run on it.

Framework Desktop 128 GB limitations

Memory is soldered LPDDR5X — unusually for Framework, this is the one component that cannot be upgraded.
ROCm/Vulkan support for Strix Halo is younger than CUDA; check your runtime supports it before buying.

Recommended setup

Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving

How these numbers are calculated

Memory need = quantized weights + KV cache at 8,192 tokens (f16) + 0.8 GB runtime overhead.
128 GB unified memory at 256 GB/s, shared between CPU and GPU.
Throughput is a memory-bandwidth roofline estimate, not a measurement. It is labelled "estimated" everywhere it appears.
KV cache is inferred from the parameter count because this model does not publish its full attention configuration.

FAQ