llama.cpp ROCm Instinct MI300X llama-2-7b Q4_0 pp512 noFA
11476.4 tok/s (pp512, batch 1)
Fields
id
benchmarks/llamacpp-rocm-mi300x-llama27b-q40-nofa-pp512Metric
prefill_tok_s
Value
11,476.4
Unit
tok/s (pp512, 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 ROCm/HIP backend). llama-bench llama-2-7b.Q4_0.gguf (TheBloke, 3.56 GiB, 6.74B params), pp512 prefill + tg128 batch-1 decode, ngl=99 full offload, ROCm. Build commit ee3a9fc, reporter @yeahdongcn. Chip Instinct MI300X (192 GB HBM3 8192-bit). Per-request batch-1, NOT aggregate.
Measured by
third_party
Reproducible
yes