lanbench -c 192.168.1.100 -p 5001 -t 30 -b 1500 -d both
In the rapidly evolving landscape of artificial intelligence, the race to build the fastest, most efficient large language model (LLM) is relentless. However, for developers, data scientists, and on-premise AI engineers, a crucial question remains: How do we measure real-world performance on our own hardware?
Most importantly, LANBench is . Whether you are running Llama 3 on a single NVIDIA RTX 4090, Mixtral on a cluster of A100s, or a quantized model on a CPU-only server, LANBench provides empirical data on how that model performs across your network .
Unidirectional tests (send or receive only) usually yield consistent results. Bidirectional Variance:
The following are the command-line options available in LANBench: