Perplexity Open-Sources TransferEngine and pplx garden for high level AI Infrastructure Optimization

Perplexity open-sourced TransferEngine and pplx garden, infrastructure tools for running trillion-parameter language models across mixed GPU clusters without vendor lock-in. The system achieves 400 Gbps throughput on both NVIDIA ConnectX-7 and AWS EFA by targeting only the intersection of guarantees across network interface controllers. Unlike alternatives (DeepEP, NVSHMEM, MoonCake) that optimize for single vendors, TransferEngine provides cross-platform portability with a minimal Rust API, enabling deployment of massive mixture-of-experts models like DeepSeek V3 and Qwen2 across heterogeneous hardware.