llama.cpp CUDA on L40 48GB: Llama-2-7B Q4_0 FA-off pp512 prefill tok/s
8870.49 tok/s (pp512, per-request, batch 1)
Fields
id
benchmarks/llamacpp-cuda-l40-llama27b-q40-nofa-pp512Metric
prefill_tok_s
Value
8,870.49
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 #15013, maintainer-curated CUDA backend). llama-bench llama-2-7b.Q4_0.gguf, ngl=99, -fa 0. L40 48GB. Build commit ee09828, reporter @Hedede. llama-bench default repeats, mean. Single device, per-request batch-1, NOT aggregate.
Measured by
third_party
Reproducible
yes
notes
Paired with the other row from the same run.
Notes
Paired with the other row from the same run.