inference/research
draftlow · 0.50updated 2026-10-17

llama.cpp discussion #4167: Apple Silicon llama-bench table

benchmark
This record is a draft. Not yet verified. Provisional. Treat every number on this page as provisional.

Fields

id
sources/llama-cpp-discussion-4167-apple-silicon-llama-bench-table
URL
https://github.com/ggml-org/llama.cpp/discussions/4167
Publisher
llama.cpp community
Kind
benchmark
Accessed
2026-10-17
notes
Re-read 2026-10-17 via the GitHub discussions API (279 comments). New rows filed from comments 16886673 (base M4 Mac mini), 18201257 (M3 Max 40-core rerun at c1d0e7a, normal and Low Power Mode), 16353087 and 16400323 (M5 Max 40-core on the pre-tensor build 8e672ef, 64 GB and 128 GB), 16556242 (M5 Pro 16-core on builds c3a1128, 073bb2c and 8e672ef), 18191340 (M1 Ultra 64-core at c1d0e7a) and 18555918 (M2 Max 30-core, Qwen3.6-35B-A3B and Qwen3.8-27B at pp4096). The pinned scoreboard cells were already filed and were not repeated.

Notes

Re-read 2026-10-17 via the GitHub discussions API (279 comments). New rows filed from comments 16886673 (base M4 Mac mini), 18201257 (M3 Max 40-core rerun at c1d0e7a, normal and Low Power Mode), 16353087 and 16400323 (M5 Max 40-core on the pre-tensor build 8e672ef, 64 GB and 128 GB), 16556242 (M5 Pro 16-core on builds c3a1128, 073bb2c and 8e672ef), 18191340 (M1 Ultra 64-core at c1d0e7a) and 18555918 (M2 Max 30-core, Qwen3.6-35B-A3B and Qwen3.8-27B at pp4096). The pinned scoreboard cells were already filed and were not repeated.

Cross-references

This record references no other records.