SGLang Summit Workshop by Crusoe: Serving SGLang on AMD and NVIDIA with Kubernetes
A SGLang Summit ticket is required to attend the event and get approved on Luma. Both Full and Expo passes are accepted.
Ticket 🎫 https://luma.com/sglang-summit
One Engine, Two Architectures: Serving SGLang on AMD and NVIDIA with Kubernetes
Deploy, tune, and benchmark SGLang across AMD and NVIDIA GPUs in this 90-minute hands-on workshop.
You’ll deploy a production-grade SGLang inference endpoint on Crusoe Managed Kubernetes using KServe, then explore how serving configurations affect performance across two GPU architectures. By benchmarking the same open-weight model on both platforms, you’ll gain practical experience building a reproducible, hardware-portable serving stack.
What You’ll Do
Deploy an inference endpoint: Run SGLang on Crusoe Managed Kubernetes with KServe.
Tune your configuration: Adjust tensor parallelism (TP), serving flags, and KV cache tiering.
Benchmark across architectures: Measure the performance of the same open-weight model on AMD and NVIDIA GPUs.
Build a reproducible stack: Leave with a serving setup you can adapt across GPU platforms.
Crusoe will provide clusters, login instructions, and access at the start of the workshop.
Who Should Join
Intermediate to advanced developers and infrastructure engineers interested in model serving, Kubernetes, and GPU performance.