Configs › llm-jp-4.1-32b-a3b-thinking › llm-jp-4.1-32b-a3b-thinking on Strix Halo: 81 tok/s, 4 measured configs
llm-jp-4.1-32b-a3b-thinking on Strix Halo: 81 tok/s, 4 measured configs
Measured llm-jp-4.1-32b-a3b-thinking speeds on Strix Halo: fastest 81 tok/s single-stream (Q4_K_M, llama.cpp/vulkan (depth-0)), 4 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 llm-jp-4.1-32b-a3b-thinking on Strix Halo: 81 tok/s (Q4_K_M, llama.cpp/vulkan (depth-0)), snapshot 2026-10-29.
Every measured single-stream run of llm-jp-4.1-32b-a3b-thinking on Strix Halo
| Decode tok/s | Quant | Backend | Context | Trust | Measured by | Source checked | Source |
|---|---|---|---|---|---|---|---|
| 81 | Q4_K_M | llama.cpp/vulkan (depth-0) | Verified (ran it) | sypherin | 2026-10-29 | source | |
| 73.6 | Q4_K_M | llama.cpp/rocm (depth-0) | Verified (ran it) | sypherin | 2026-10-29 | source | |
| 72.6 | Q4_K_M | llama.cpp/vulkan (depth-4096) | 4096 | Verified (ran it) | sypherin | 2026-10-29 | source |
| 64.8 | Q4_K_M | llama.cpp/rocm (depth-4096) | 4096 | Verified (ran it) | sypherin | 2026-10-29 | source |
Backends on Strix Halo
Fastest plain-decoding run per backend. Quants can differ between rows.
- llama.cpp/vulkan: 81 tok/s (Q4_K_M)
- llama.cpp/rocm: 73.6 tok/s (Q4_K_M)
Questions and answers
- How fast is llm-jp-4.1-32b-a3b-thinking on Strix Halo?
- The fastest of 4 measured configs is 81 tok/s single-stream (Q4_K_M, llama.cpp/vulkan (depth-0)). The slowest is 64.8 tok/s.
- Which backend is fastest for llm-jp-4.1-32b-a3b-thinking on Strix Halo?
- llama.cpp/vulkan at 81 tok/s, then llama.cpp/rocm at 73.6. Each backend figure is its best plain-decoding row and may use a different quant.
See also: llm-jp-4.1-32b-a3b-thinking 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.