Geerling llama.cpp Vulkan on rtxa4000 (pc host): gpt-oss-20b Q4_K_M pp512 3182.40 tok/s
3182.4 tok/s (pp512)
Fields
id
benchmarks/geerling-vulkan-rtxa4000-pc-gpt-oss-20b-q4km-pp512Metric
prefill_tok_s
Value
3,182.4
Unit
tok/s (pp512)
Model
gpt-oss-20b-Q4_K_M.gguf (10.81 GiB, 20.91 B params)
Methodology
THIRD-PARTY MEASURED (Jeff Geerling, geerlingguy/ai-benchmarks). llama.cpp build not printed, backend Vulkan, default llama-bench flags (ngl 99). Host: Intel Core Ultra 265K, 64GB DDR5, Ubuntu 25.10. Model file size 10.81 GiB, 20.91 B params, Q4_K_M. prefill tok/s, prompt length 512, per-request, batch 1. NOT aggregate. Whole-system wall power during the run 297.2 W (node scope, not device).
Measured by
third_party
Reproducible
yes
Mean power (W)
297.2
Power scope
node
Thermal state
unknown
notes
Source issue prints the llama-bench mean and the +/- spread; the stored value is the mean. The Pi-host rows are the same physical GPU on a PCIe-constrained host, so they are not a card ceiling.
Notes
Source issue prints the llama-bench mean and the +/- spread; the stored value is the mean. The Pi-host rows are the same physical GPU on a PCIe-constrained host, so they are not a card ceiling.