Gufo on the AMD Strix Halo 128GB
On the AMD Strix Halo 128GB, Gufo runs Qwen3.8 125B · 6B active at 4-bit at a median 26.0 tokens per second for generation, from 3 plain runs (132K to 134K context, 1 not stated).
Speed by model on the AMD Strix Halo 128GB
Plain runs only: one device, one request, no speculative decoding, the whole model in memory. Each row is one model size at one bit class, with the other engines' plain runs of the same model on this card beside it. Context length and build still differ between runs, and each side shows its own.
| Model | Gufo | Other engines |
|---|---|---|
| Qwen3.8 125B · 6B active 4-bit | 26.0 t/smedian of 3, 132K to 134K context, 1 not stated | llama.cpp 23.0 t/smedian of 3, 128K to 160K context, 1 not stated |
| Gemma 4 26B · 4B active 4-bit | 67.3 t/smedian of 2, context not stated | no other engine reported |
| Gemma 4 31B 4-bit | 11.5 t/smedian of 2, context not stated | no other engine reported |
| Gemma 4 31B 8-bit | 6.4 t/smedian of 2, context not stated | no other engine reported |
| Qwen3.6 35B · 3B active 8-bit | 50.8 t/sone report, context not stated | llama.cpp 24.1 t/sone report, 512K context, 1 not stated |
33 Gufo reports on the AMD Strix Halo 128GB
Newest first. A run on several cards, with speculative decoding, with part of the model in system RAM or with batched requests says so, since each describes a different setup.
- Qwen3.8 125B · 6B active · UD-Q4_K_XL54.1 t/s(speculative decoding)
- Gemma 4 26B · 4B active · NVFP4114.6 t/s
- Gemma 4 31B · UD-Q8_K_XL6.3 t/s
- Qwen3.8 125B · 6B active · UD-IQ4_XSPP 1821.0 t/s
- Gemma 4 26B · 4B active · Q4_K_XL53.2 t/s
- Qwen3.8 · UD-Q4_K_XL71.1 t/s(speculative decoding)
- Qwen3.8 · UD-Q4_K_XL39.9 t/s(speculative decoding)
- Gemma 4 26B · 4B active · QAT UD-Q4_K_XL (Q4_0)81.4 t/s
- Gemma 4 31B · QAT UD-Q4_K_XL (Q4_0)12.1 t/s
- Gemma 4 31B · UD-Q8_K_XL6.4 t/s
- Gemma 4 31B · UD-Q4_K_XL10.9 t/s
- Qwen3.8 125B · 6B active · Q4_K_XL60.4 t/s
- Qwen3.8 125B · 6B active · Q8_029.8 t/s(2 cards)
- Gemma 4 26B · 4B active · UD-Q4_K_XL53.2 t/s
- Qwen3.8 125B · 6B active · Q4 Mix (Q4_K/Q5_1/IQ4_NL/Q8_0)36.9 t/s(speculative decoding)
- Qwen3.8 125B · 6B active · UD-Q4_K_XLPP 1057.5 t/s
- Qwen3.8 125B · 6B active · UD-Q4_K_XL34.2 t/s(2 cards)
- DeepSeek V4 Flash 284B · 13B active · IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ816.3 t/s(part in system RAM)
- Qwen3.8 125B · 6B active · UD-Q4_K_XL21.1 t/s
- Qwen3.8 125B · 6B active · UD-Q4_K_XL25.9 t/s
- Qwen3.8 27B10.0 t/s
- Qwen3.8 125B · 6B active · UD-Q4_K_XLPP 873.0 t/s(speculative decoding)
- Qwen3.8 125B · 6B active46.0 t/s
- Qwen3.8 125B · 6B active · UD-IQ4_XS50.0 t/s(speculative decoding)
- Qwen3.8 27B · Q4_K_XL34.0 t/s(speculative decoding)
- Qwen3.8 125B · 6B active · UD-Q4_K_XLPP 1124.0 t/s(speculative decoding)
- Qwen3.8 125B · 6B active · UD_Q4_XL40.0 t/s
- Qwen3.6 35B · 3B active · Q8_K_XL50.8 t/s
- Qwen3.8 125B · 6B active · UD-Q4_K_XL33.6 t/s(speculative decoding)
- Qwen3.8 125B · 6B active · UD-Q4_K_XL32.9 t/s(speculative decoding)
Frequently asked
How fast is Gufo on the AMD Strix Halo 128GB?
On the AMD Strix Halo 128GB, Gufo runs Qwen3.8 125B · 6B active at 4-bit at a median 26.0 tokens per second for generation, from 3 plain runs (132K to 134K context, 1 not stated).
Is Gufo faster than other engines on the AMD Strix Halo 128GB?
Same card, same model size and bit class, plain runs only: Qwen3.8 125B · 6B active at 4-bit gives Gufo 26.0 t/s (3 runs) and llama.cpp 23.0 t/s (3). Context length and engine build differ between the runs, so treat it as a lead, not a benchmark.