Can the Ryzen AI Max+ 395 (128 GB) run Qwen3.8-Flash-Next?
Qwen3.8-Flash-Next·180B (61.6B active)·Ryzen AI Max+ 395 (128 GB)·128 GB unified memory
Yes, but only at Q3_K_M (about 87 GB of 96 GB), which costs noticeable quality, at roughly 4.7 tokens/s.
- best quant
- Q3_K_M
- needs
- 87 GB
- usable
- 96 GB
- est. speed
- ~4.7 tok/s
Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the Qwen3.8-Flash-Next page →
Every quantisation of Qwen3.8-Flash-Next on the Ryzen AI Max+ 395 (128 GB)
| Quant | Memory | Fit | Speed |
|---|---|---|---|
| Q2_K | 59 GB | fits | ~7 tok/s |
| Q3_K_M pick | 87 GB | fits | ~4.7 tok/s |
| Q4_K_M | 108 GB | too big | — |
| Q5_K_M | 126 GB | too big | — |
| Q6_K | 146 GB | too big | — |
| Q8_0 | 185 GB | too big | — |
| FP16 / BF16 | 347 GB | too big | — |
Speed is a bandwidth-based estimate for generating text (256 GB/s); prompt processing and long contexts are slower.
Measured speed on the Ryzen AI Max+ 395 (128 GB)
No community reports for this pair yet.
Report your own speed →Run Qwen3.8-Flash-Next on the Ryzen AI Max+ 395 (128 GB)
Using Q3_K_M, the best fit for Ryzen AI Max+ 395 (128 GB).
ollama run qwen3.8-flash-nextThe library build picks its own default quantisation.
ollama run hf.co/unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q3_K_XLdocker run -d --name ollama --device /dev/kfd --device /dev/dri \
-v ollama:/root/.ollama -p 11434:11434 ollama/ollama:rocm
docker exec -it ollama ollama run qwen3.8-flash-nextThe image bundles the ROCm runtime; the host only needs the amdgpu kernel driver.
llama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q3_K_XLNeeds a HIP (ROCm) build of llama.cpp. Downloads the GGUF on first run.
docker run -d --name llama --device /dev/kfd --device /dev/dri -p 8080:8080 \
-v ~/.cache/llama.cpp:/root/.cache/llama.cpp \
ghcr.io/ggml-org/llama.cpp:server-rocm \
-hf unsloth/Qwen3.8-Flash-Next-GGUF:UD-Q3_K_XL --host 0.0.0.0 --port 8080 -ngl 99Server and web UI on port 8080.
lms get unsloth/Qwen3.8-Flash-Next-GGUF@UD-Q3_K_XLThen load it from the LM Studio app or with `lms load`.
vllm serve Qwen/Qwen3.8-Flash-NextNeeds vLLM built for ROCm; it supports Instinct MI200/MI300 and Radeon RX 7900 / 9000 cards.
vllm serve Qwen/Qwen3.8-Flash-Next-FP8services:
vllm:
image: rocm/vllm:latest
command: vllm serve Qwen/Qwen3.8-Flash-Next --host 0.0.0.0 --port 8000
ports:
- "8000:8000"
ipc: host
devices:
- /dev/kfd
- /dev/dri
group_add:
- video
security_opt:
- seccomp=unconfined
volumes:
- ~/.cache/huggingface:/root/.cache/huggingfaceAMD’s rocm/vllm image. API on port 8000.
sglang serve Qwen/Qwen3.8-Flash-Next --host 0.0.0.0 --port 30000Needs SGLang’s ROCm build (Instinct MI300-class GPUs).
from transformers import pipeline
pipe = pipeline("image-text-to-text", model="Qwen/Qwen3.8-Flash-Next", dtype="auto", device_map="auto")
messages = [
{
"role": "user",
"content": [
{"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
{"type": "text", "text": "Describe this image."},
],
}
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])Vision models that fit the Ryzen AI Max+ 395 (128 GB) comfortably
All that fit →| Model | Params | Best quant | Memory | Speed |
|---|---|---|---|---|
| 125BA10B | Q5_K_M | ≈91 GBFits in memory | ~20 tok/s | |
| 124BA12B | Q5_K_M | ≈88 GBFits in memory | ~16 tok/s | |
| 109BA18B | Q6_K | ≈88 GBFits in memory | ~9.5 tok/s | |
| 108BA14.3B | Q5_K_M | ≈81 GBFits in memory | ~14 tok/s | |
| 73.4B | Q8_0 | ≈79 GBFits in memory | ~1.8 tok/s | |
| 36BA3B | Q8_0 | ≈38 GBFits in memory | ~44 tok/s |
Other GPUs that run Qwen3.8-Flash-Next
Frequently asked questions
- Can the Ryzen AI Max+ 395 (128 GB) run Qwen3.8-Flash-Next?
- Yes, but only at Q3_K_M (about 87 GB of 96 GB), which costs noticeable quality, at roughly 4.7 tokens/s.
- How fast does Qwen3.8-Flash-Next run on the Ryzen AI Max+ 395 (128 GB)?
- An estimate from memory bandwidth gives about 4.7 tokens/s at Q3_K_M.
- Which quantisation of Qwen3.8-Flash-Next should I use on the Ryzen AI Max+ 395 (128 GB)?
- Q3_K_M: the largest one that fits. Expected quality: noticeable loss compared with the full-precision weights.
Memory and speed are estimates from the model's size and the card's memory bandwidth; real numbers depend on the runtime, drivers and context length. Community figures are medians of user reports.