llama.cpp Vulkan NVIDIA TITAN V llama-2-7b Q4_0 pp512 FA
780.48 tok/s (pp512)
Fields
id
benchmarks/llamacpp-vulkan-titanv-llama27b-q40-fa-pp512Metric
prefill_tok_s
Value
780.48
Unit
tok/s (pp512)
Model
llama-2-7b.Q4_0.gguf (TheBloke, 3.56 GiB, 6.74B params)
Methodology
COMMUNITY-SCOREBOARD (llama.cpp discussion #10879, maintainer-curated Vulkan backend). llama-bench llama-2-7b.Q4_0.gguf (TheBloke, 3.56 GiB, 6.74B params), pp512 and tg128, batch 1, ngl=99 full offload, Vulkan, flash attention on. Build e56abd2. Chip NVIDIA TITAN V. Per-request batch-1, not aggregate. Printed mean 780.48 +/- 3.60.
Measured by
third_party
Reproducible
yes
notes
Flash-attention-on scoreboard row.
Notes
Flash-attention-on scoreboard row.