llama.cpp ROCm on Instinct MI210: Llama-2-7B Q4_0 FA-off pp512 prefill tok/s (scoreboard #15021)
2486.22 tok/s (pp512, per-request, batch 1)
Fields
id
benchmarks/llamacpp-rocm-mi210-llama27b-q40-nofa-pp512Metric
prefill_tok_s
Value
2,486.22
Unit
tok/s (pp512, per-request, batch 1)
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. Instinct MI210 (64 GB HBM2e 4096-bit). Build commit 8160b38, reporter @65a. llama-bench default -r 5, mean +/- std. Single-GPU per thread rules. Per-request batch-1, NOT aggregate.
Measured by
third_party
Reproducible
yes
notes
Maintainer-verified scoreboard row from ggml-org/llama.cpp#15021 (ROCm/HIP backend). Non-FA baseline.
Notes
Maintainer-verified scoreboard row from ggml-org/llama.cpp#15021 (ROCm/HIP backend). Non-FA baseline.