RTX 4070 Ti SUPER llama.cpp: gpt-oss 20B MXFP4 prefill at 16k ctx (Hardware Corner)
4182.7 tok/s prompt processing, batch 1, per-request, ctx 16k
Fields
id
benchmarks/hc-4070tis-gptoss-20b-mxfp4-16k-ppMetric
prefill_tok_s
Value
4,182.7
Unit
tok/s prompt processing, batch 1, per-request, ctx 16k
Model
gpt-oss 20B
Methodology
Hardware Corner per-GPU context-tier table (March 2026 update), single RTX 4070 Ti SUPER, llama.cpp GGUF, batch-1 per-request. gpt-oss 20B MXFP4, 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 already exists as a decode record at this context, not duplicated. Context is the tier the page groups into, not a llama-bench -d argument.
Notes
Paired generation cell already exists as a decode record at this context, not duplicated. Context is the tier the page groups into, not a llama-bench -d argument.