Hardware Corner RTX 3060 12GB local LLM benchmark page (March 2026)

Per-SKU llama.cpp table, page header 'Updated Benchmarks: March 2026'. Single RTX 3060 12GB (12 GB GDDR6, 360 GB/s stated). Tables: Qwen3 8B Q4_K at 4k/16k/32k (PP 1,696.8 / 1,119.2 / 764.7; TG 55.2 / 42.0 / 31.9) and Qwen3 14B Q4_K at 4k/16k (PP 972.6 / 678.2; TG 31.2 / 22.7). Header '22.6 t/s / 678 t/s on 14B at 16k' matches the 16k cells rounded, while the header also labels the same numbers '14B @ 4k'. Key insights claim Qwen3 8B up to 64k, but the published table stops at 32k and that deeper column was not invented. No build hash. Distinct from the earlier two-row hardware-corner ingest of this card, which covered a different model/context pair.