RadixArk and Google build a TPUv7 megakernel for Muse Spark Open

AISGLang says RadixArk is working with Google on a fully pipelined megakernel for Muse Spark Open, optimized for long-context agentic workloads. SGLang says performance on TPUv7 is now very competitive with popular GPU options.

SGLangOfficial

@sgl_project

View on X

Congrats to the Google TPU team, and thanks @SemiAnalysis_ for a clear look at where TPU stands today and the potential it holds for production agentic workloads.

We’re excited to see @radixark work with Google to build a fully pipelined megakernel for Muse Spark Open, with deep optimizations for long-context agentic workloads. On TPUv7, the performance is now very competitive with popular GPU options.

More models on SGLang-TorchTPU are in the works, and TPU will only get stronger for anyone serving AI at scale 🧡

SemiAnalysis@SemiAnalysis_

The Iron(wood) Throne: Google's TPU Stakes Its Claim on Inference Viable Alternative of CUDA Moat, NVIDIA Working Nights & Weekends, Jensen’s Counter Submission, Kimi K3, Meta Muse Spark Open, GLM5.3, TorchTPU, SGLang, vLLM, llm-d, Ray https://newsletter.semianalysis.com/p/tpu-strikes-the-king

View quoted post on X

Source details

SGLang · x.comPublished Nov 5, 6:08 PM ET

Original title: Congrats to the Google TPU team, and thanks @SemiAnalysis_ for a clear look at where TPU stands today and the potential it holds for prod...

RadixArk and Google build a TPUv7 megakernel for Muse Spark Open

Topics