Strix Halo + Radeon AI Pro R9700 over OCuLink: Qwen3.8 Flash-Next at ~102 tok/s (custom engine)
| GPU(s) | AMD Radeon AI Pro R9700 32 GB (OCuLink x4) + Ryzen AI Max+ 395 Radeon 8060S iGPU Amazon · eBay (used) · Newegg |
|---|---|
| VRAM / unified memory (GB) | 160 |
| CPU | AMD Ryzen AI Max+ (Strix Halo) mini PC Amazon · eBay (used) · Newegg |
| Motherboard | |
| RAM | 128 GB unified (Strix Halo) |
| PSU | |
| Other | Lian Li B4 case; R9700 power-capped to 210 W, -500 MHz, fan capped at 1500 rpm; iGPU capped at 2000 MHz, CPU boost off |
| Reported total cost | |
| Power | R9700 capped at 210 W |
Measured speed (generation tokens/sec)
📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
| Model | Quant | Engine | tok/s | Details |
|---|---|---|---|---|
| Qwen3.8 Flash-Next | custom int4 experts / 6-bit trunk | strixpro (MTP depth 3 + prompt lookup) | 102 | ⚡ spec-decode (MTP) |
| Qwen3.8 Flash-Next | custom int4 experts / 6-bit trunk | strixpro, ~100K context | 82 | |
| Qwen3.8 Flash-Next | custom int4 experts / 6-bit trunk | strixpro, ~980K context | 71 | |
| Qwen3.8 Flash-Next | GGUF | llama.cpp (Strix Halo iGPU only) | 29-34 |
Blog post 5 Oct 2026, updated 16 Oct 2026. Hot experts and trunk run on the R9700, all experts held on the Strix Halo. Prefill about 2,000 tok/s on 4-16K prompts. "~102" is the author's approximate figure for short prompts.
Source: https://vuln.dev/flash-engine/
Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.
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)