geerlingguy/ai-benchmarks issue #40 — RTX 3060 12GB llama.cpp Vulkan llama-bench (build 933414c0b)

Fields

id

sources/geerling-ai-benchmarks-issue-40-rtx3060

URL

https://github.com/geerlingguy/ai-benchmarks/issues/40

Publisher

Jeff Geerling

Kind

benchmark

Published

2025-12-06

Accessed

2026-10-17

notes

Jeff Geerling llama-bench. Hosts: Raspberry Pi CM5 16GB (Pi OS Trixie) and Intel Core Ultra 265K / 64GB Ubuntu 25.10. Vulkan, ngl 99, build 933414c0b (7293). Models Q4_K_M: tinyllama 1.1B, Llama-3.2-3B, llama-2-7b, llama-2-13b, DeepSeek-R1-Distill-Qwen-14B, gpt-oss-20b. Default tests pp512, pp4096, tg128, pp4096+tg128. gpt-oss-20b rerun pp512+tg128 after ErrorOutOfDeviceMemory. Power is whole-system wall draw.

Notes

Jeff Geerling llama-bench. Hosts: Raspberry Pi CM5 16GB (Pi OS Trixie) and Intel Core Ultra 265K / 64GB Ubuntu 25.10. Vulkan, ngl 99, build 933414c0b (7293). Models Q4_K_M: tinyllama 1.1B, Llama-3.2-3B, llama-2-7b, llama-2-13b, DeepSeek-R1-Distill-Qwen-14B, gpt-oss-20b. Default tests pp512, pp4096, tg128, pp4096+tg128. gpt-oss-20b rerun pp512+tg128 after ErrorOutOfDeviceMemory. Power is whole-system wall draw.

Cross-references

This record references no other records.