llama.cpp ROCm on AI Max+ 395 (Strix Halo): Llama-2-7B Q4_0 noFA pp512 prefill tok/s (scoreboard #15021)
911.36 tok/s (pp512, batch 512)
Fields
id
benchmarks/llamacpp-rocm-8060s-run1-llama27b-q40-nofa-pp512Metric
prefill_tok_s
Value
911.36
Unit
tok/s (pp512, batch 512)
Model
llama-2-7b.Q4_0.gguf (TheBloke, 3.56 GiB, 6.74B params)
Methodology
COMMUNITY-SCOREBOARD (llama.cpp discussion #15021, maintainer-curated). llama-bench -m llama-2-7b.Q4_0.gguf -ngl 99 -fa 0, ROCm/HIP backend. Build commit e60f241, reporter @firefox42. Default -r 5 repeats, mean +/- std. Single GPU per run. Per-request batch-1, NOT aggregate. Hardware row: 128 GB / LPDDR5.
Measured by
third_party
Reproducible
yes
notes
Maintainer-verified scoreboard row from ggml-org/llama.cpp#15021 (ROCm). Identical model/test/flags across every AMD GPU.
Notes
Maintainer-verified scoreboard row from ggml-org/llama.cpp#15021 (ROCm). Identical model/test/flags across every AMD GPU.