| # 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 |
