RTX 5070 llama.cpp: Qwen3 14B Q4_K prefill at 16k ctx (Hardware Corner)
1315.2 tok/s prompt processing, batch 1, per-request, ctx 16k
Fields
id
benchmarks/hc-5070-qwen3-14b-q4k-16k-ppMetric
prefill_tok_s
Value
1,315.2
Unit
tok/s prompt processing, batch 1, per-request, ctx 16k
Model
Qwen3 14B
Methodology
Hardware Corner per-GPU context-tier table (March 2026 update), single RTX 5070, llama.cpp GGUF, batch-1 per-request. Qwen3 14B Q4_K, context depth 16k. Prefill column is prompt-processing tok/s; generation column is decode tok/s. NOT an aggregate across concurrent requests. The page does not reprint the llama-bench build hash (the site's sibling ranking page pins CUDA 12.8 / Q4_K_XL).
Measured by
third_party
Reproducible
yes
notes
Paired generation cell is recorded separately. Context is the tier the page groups into, not a llama-bench -d argument.
Notes
Paired generation cell is recorded separately. Context is the tier the page groups into, not a llama-bench -d argument.