Database Mart Ollama 0.13.5 benchmark, RTX PRO 6000 Blackwell

Database Mart hosted-GPU report, page footer dated 10/07/2026. One NVIDIA RTX PRO 6000 Blackwell, 96 GB GDDR7, compute capability 12.0, 24,064 CUDA cores. Engine Ollama 0.13.5 (llama.cpp backend; the llama.cpp commit is not printed). Quantization is the default Ollama pull, labelled 4-bit, exact GGUF quant not printed, so format_id is null. Prompt is the single question 'What is GPU?' with prompt token counts of 6 to 70, so the prompt-eval column is NOT recorded as prefill. Output length is whatever the model produced (356 to 2,729 tokens), not a fixed generation length. Eval rate is per-request decode, batch size 1, NOT aggregate. Distinct from the Hardware Corner llama-bench tables already filed.