llmBench maps local GPU perf to Arena leaderboard

The gap between "can I run it?" and "how good is it globally?" has been a real frustration for local LLM enthusiasts — llmBench is a direct attempt to bridge both questions in one tool.

• –Unique angle: maps local tokens/sec and VRAM metrics against the LMSYS Arena leaderboard, giving consumer hardware a global performance frame of reference that tools like LocalScore don't provide
• –Tracks energy efficiency (Joules per token) and thermal behavior alongside standard throughput — useful for laptop users running on constrained TDPs
• –Hardware forensic mode digs into PCIe bandwidth, RAM manufacturer, and DDR generation — surfacing hidden bottlenecks beyond just VRAM size
• –Currently Windows-only (WMI-dependent) and requires an NVIDIA GPU with nvidia-smi, limiting the audience
• –Very early stage (2 GitHub stars, 8 commits) — promising concept but not yet battle-tested or cross-platform