# Poor Paul's Benchmark Results Dataset
# https://huggingface.co/datasets/paulplee/ppb-results

## What This Is
Community-submitted GPU benchmark results for local LLM inference.
Schema: 61 columns covering hardware, model, quantization, throughput, TTFT, ITL, power, and thermals.

## How to Query
MCP server: mcp.poorpaul.dev (tools: recommend_quantization, query_ppb_results, get_gpu_headroom, list_tested_configs)
Analytics:  poorpaul.dev/insights

## How to Contribute
Tool: https://github.com/paulplee/poor-pauls-benchmark
Run: ppb all suites/my_gpu.toml

## Key Fields
gpu_name, gpu_vram_gb, quant, throughput_tok_s, avg_ttft_ms, avg_itl_ms, concurrent_users, avg_power_w