Configs › gemma-4-26B-A4B-it › gemma-4-26B-A4B-it on Strix Halo: 78 tok/s, 5 measured configs
gemma-4-26B-A4B-it on Strix Halo: 78 tok/s, 5 measured configs
Measured gemma-4-26B-A4B-it speeds on Strix Halo: fastest 78 tok/s single-stream (UD-Q4_K_XL, llama.cpp), 5 configs with quant, backend and context, each linked to its source, snapshot 2026-10-29.
Snapshot 2026-10-29 · 413 configs on AI boxes and 288 on GPU cards · 121 models. Every number is generated from the tracker snapshot and linked to where it was measured; nothing here is typed by hand.
Fastest measured single-stream run of gemma-4-26B-A4B-it on Strix Halo: 78 tok/s (UD-Q4_K_XL, llama.cpp), snapshot 2026-10-29.
Every measured single-stream run of gemma-4-26B-A4B-it on Strix Halo
| Decode tok/s | Quant | Backend | Context | Trust | Measured by | Source checked | Source |
|---|---|---|---|---|---|---|---|
| 78 | UD-Q4_K_XL | llama.cpp | Owner-submitted | sypherin | 2026-09-16 | source | |
| 65.8 | Q4_0 | llama.cpp/vulkan | Extracted, source-linked | huppiflupp | 2026-10-24 | source | |
| 54.2 | Q4_K_M | llama.cpp/vulkan | 512 | Structured table | hogeheer | 2026-10-14 | source |
| 48.5 | Q4_K_M | llama.cpp/vulkan | 512 | Structured table | hogeheer | 2026-10-14 | source |
| 41.8 | UD-Q8_K_XL | llama.cpp/rocm-7.2.4 | Structured table | kyuz0 | 2026-10-29 | source |
Best quant on Strix Halo
Fastest plain-decoding run per quant, versus the fastest quant.
- UD-Q4_K_XL: 78 tok/s, fastest (llama.cpp)
- Q4_0: 65.8 tok/s, 16% lower (llama.cpp/vulkan)
- Q4_K_M: 54.2 tok/s, 31% lower (llama.cpp/vulkan)
- UD-Q8_K_XL: 41.8 tok/s, 46% lower (llama.cpp/rocm-7.2.4)
Backends on Strix Halo
Fastest plain-decoding run per backend. Quants can differ between rows.
- llama.cpp: 78 tok/s (UD-Q4_K_XL)
- llama.cpp/vulkan: 65.8 tok/s (Q4_0)
- llama.cpp/rocm-7.2.4: 41.8 tok/s (UD-Q8_K_XL)
gemma-4-26B-A4B-it on other boxes
- DGX Spark: 69.9 tok/s best of 8 configs
Questions and answers
- How fast is gemma-4-26B-A4B-it on Strix Halo?
- The fastest of 5 measured configs is 78 tok/s single-stream (UD-Q4_K_XL, llama.cpp). The slowest is 41.8 tok/s.
- Which quant of gemma-4-26B-A4B-it is fastest on Strix Halo?
- UD-Q4_K_XL at 78 tok/s, ahead of Q4_0 at 65.8, Q4_K_M at 54.2, UD-Q8_K_XL at 41.8. Plain decoding only.
- Which backend is fastest for gemma-4-26B-A4B-it on Strix Halo?
- llama.cpp at 78 tok/s, then llama.cpp/vulkan at 65.8, llama.cpp/rocm-7.2.4 at 41.8. Each backend figure is its best plain-decoding row and may use a different quant.
- How does Strix Halo compare with other boxes for gemma-4-26B-A4B-it?
- Strix Halo best 78 tok/s. Other boxes, best measured: DGX Spark 69.9. Backends and quants differ, so this is the best known number per box, not a controlled test.
See also: gemma-4-26B-A4B-it on every box · Strix Halo hardware page (from S$4,329 in Singapore today).
Last updated: snapshot 2026-10-29; newest source check 2026-10-29.
More: all configs · hardware · news · contributors · methodology · llms.txt · llms-full.txt · API (OpenAPI) · JSON snapshot. Built by Altronis, private on-prem AI, Singapore.