RTX 5070 12GB + 32 GB DDR4: Qwen3.8-Flash-Next 177B at 11.5 tok/s with expert streaming
| GPU(s) | RTX 5070 12GB Amazon · eBay (used) · Newegg |
|---|---|
| VRAM / unified memory (GB) | 12 |
| CPU | AMD Ryzen 5 5600GT Amazon · eBay (used) · Newegg |
| Motherboard | |
| RAM | 32 GB DDR4-2400 Amazon · eBay (used) · Newegg |
| PSU | |
| Other | PCIe Gen3, Windows, custom llama.cpp-based expert streaming with a page-locked hot-expert tier |
| Reported total cost | |
| Power |
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 | UD-IQ3_XXS, experts streamed from RAM/SSD | llama.cpp-based expert streaming (custom) | 11.5 | |
| Qwen3.8 Flash-Next | UD-IQ3_XXS, long coding prompt (4,892 tokens out) | llama.cpp-based expert streaming (custom) | 10.15 |
llama.cpp Show and tell discussion, 4 Oct 2026. Benchmark ~11.5 tok/s (up from ~7 on the inherited setup); the author saw 14-15 tok/s in normal conversations. Most of the model lives off the GPU (offload), so speed depends heavily on RAM/SSD I/O.
Source: https://github.com/ggml-org/llama.cpp/discussions/29930
https://github.com/MohammadHaishemKhawaja/hashyy
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)