Hardware Corner RTX 5060 Ti 16GB local LLM benchmark page (March 2026)

Per-SKU llama.cpp table, page header 'Updated Benchmarks: March 2026'. Single RTX 5060 Ti 16GB (16 GB GDDR7, 448 GB/s stated). Tables: Qwen3 8B Q4_K at 4k/16k/32k/64k (TG 72.6 / 53.2 / 38.9 / 20.7; PP 2,987.4 / 1,812.6 / 1,098.5 / 650.3), Qwen3 14B Q4_K at 4k/16k/32k (TG 40.1 / 32.9 / 26.4; PP 1,704.8 / 943.2 / 612.7), and gpt-oss 20B MXFP4 at 4k through 128k (TG 98.4 / 89.1 / 78.6 / 62.3 / 44.8; PP 4,215.6 / 3,102.4 / 2,248.9 / 1,486.2 / 892.5). Header '32.9 t/s / 943 t/s on 14B at 16k' matches the 16k cells. No 32B row exists on this page despite the 16 GB. No build hash, CUDA version, or exact prompt and output lengths inside the context bucket.