Hardware Corner RTX A6000 local LLM benchmark page (March 2026)

Per-SKU llama.cpp table, 'Updated Benchmarks: March 2026'. Single RTX A6000 (48 GB GDDR6, 768 GB/s). Tables for Qwen3 8B/14B/32B, Qwen3 30B-A3B, gpt-oss 20B MXFP4, Llama 3.3 70B, each at 4k through the largest context that fit. Token-generation and prompt-processing columns are per-request, not aggregate. The 14B/16k decode cell (40.7 t/s) matches the Dec 2025 ranking row already stored as hwcorner-rtx-a6000-qwen3-14b-q4kxl-decode-16k (40.69) within rounding, so this page is the same lab, not a second lab.