inference/research
verifiedmedium · 0.70updated 2026-10-14

kyuz0 amd-strix-halo-toolboxes results.json — 554 llama-bench pp512/tg128 runs, 5 backend envs, builds 9187/9193, Ryzen AI Max+ 395 (gfx1151)

benchmark

Fields

id
sources/kyuz0-strix-halo-toolbox-results-json
Publisher
kyuz0 / AMD Strix Halo toolboxes
Kind
benchmark
Accessed
2026-10-14
notes
Machine-run results.json powering the interactive viewer. Each run: model_clean, env (rocm-7_2_3 / rocm6_4_4 / rocm7-nightlies / vulkan_amdvlk / vulkan_radv), fa=1, ngl=99, mmap=0, pp512 or tg128 mean+std t/s. Fedora 43, kernel 6.19.12. Per-request batch-1.

Notes

Machine-run results.json powering the interactive viewer. Each run: model_clean, env (rocm-7_2_3 / rocm6_4_4 / rocm7-nightlies / vulkan_amdvlk / vulkan_radv), fa=1, ngl=99, mmap=0, pp512 or tg128 mean+std t/s. Fedora 43, kernel 6.19.12. Per-request batch-1.

Cross-references

This record references no other records.