Qwen模型在Steam Machine上解码约18 token/s
AI 总结
Reddit 用户在配备 8 GB AMD Navi 33 GPU 的 Valve Steam Machine 上,通过 Strata 运行 Qwen3.8-Flash-Next Gyro S,报告解码速度约 18 token/s、预填充速度约 235 token/s。将常驻预算设为 64 GB 后,解码速度提高约 20%;手动加载 MTP 推测解码层则未带来收益,预填充速度反而下降约 10%。数据来自该用户单台机器上每组配置三次运行,尚未经独立验证。
深度见解
该报告提供了在 8 GB 消费级 GPU 上运行大模型的详细配置与调优参考,但结果仅适用于测试者的硬件和软件环境。
I followed up on the post about running Qwen 3.8 Flash Next with 8 GB VRAM and 32 GB RAM. I wanted to see how it does on my Steam Machine.
Setup
- Valve Steam Machine, stock SteamOS 3.9.5, AMD Navi 33 with 8 GB, 32 GB RAM, NVMe
- Stock Strata terminal 0.2.9 (engine 0.1.55, HIP package), Gyro S TQ1_0
- 128K context, q4_0 KV, no vision, MTP off, served on the LAN
Results (Strata's own timing line from the engine log, about 8K-token prompts, 3 runs each)
| Config | Prefill t/s | Decode t/s |
|---|---|---|
Default (--tune skip) | 219-240 | 15.1-15.9 |
| Quick tuner | 237-239 | 14.5-17.2 (tuner said "already fastest") |
pcie-frac 0.35 | 239 | 14.8-16.0 |
kv-resident 20480 + vram-reserve-mib 900 | 227-229 | 14.8-15.5 |
resident-budget-gib 64 | 232-242 | 17.3-19.8 |
#Valve#Steam Machine#SteamOS#Qwen3.8-Flash-Next Gyro S#Strata#AMD Navi 33