At NVIDIA GTC 2025, Astera Labs is showcasing its collaboration with NVIDIA and Wistron. This is the first of many solutions using our Scorpio Smart Fabric Switches to deliver leading performance through a modular design that scales across different configurations for PCIe 6-ready NVIDIA Blackwell-based MGX platforms.
For this demo, utilizing the PCIe 6-ready NVIDIA Blackwell-based MGX platform with the Scorpio switch board, we are running chatbots as an inferencing workload on the Llama 3.3 model with 70 billion parameters. See how four independent chatbots are running on the inferencing server, ready to answer users’ questions. While running the chatbots, we can see the NVIDIA GPUs are being fully utilized showing scalable performance from 1 to 4 GPUs running in parallel.
