#TPU
8 related articles
Google Open-Sources TorchTPU: A Native PyTorch Backend Bridges the TPU Ecosystem
Google open-sources TorchTPU, providing a native PyTorch backend for TPU. This article analyzes how it differs from the XLA route, its potential impact on the CUDA ecosystem, and the costs and performance issues developers should watch during migration.
Google Open-Sources TorchTPU: How a PyTorch Native Backend Reshapes the TPU Ecosystem
Google and Meta open-source TorchTPU, a PyTorch native backend for Cloud TPU with eager execution by default and torch.compile/XLA support, set to replace PyTorch/XLA for new workloads. This article analyzes its architecture, execution trade-offs, and impact on developers and the TPU ecosystem.
TorchTPU Released: Native PyTorch Support for Google TPU
Google and Meta jointly release TorchTPU, providing a native TPU backend for PyTorch so models run out of the box. This article examines the open-source collaboration's technical positioning, impact on developer migration costs, and performance and governance questions still to be verified.
First TPUv7 Submission to AgentX: vLLM Helps Google's TPU Ecosystem Score Another Win
AgentX receives its first TPUv7-based model submission deployed via vLLM, a three-way collaboration between Google TPU, Red Hat AI, and inferact. We break down what it means for vLLM and TPU ecosystem compatibility.
Torch-TPU: Running SGLang and vLLM Natively on TPUs
Google & Meta introduce Torch-TPU at PyTorch Conference: a native PyTorch TPU backend enabling SGLang and vLLM to run on TPUs with zero rewrites to schedulers, batching, or server APIs.
Google Project Suncatcher: TPU Chips Head to Space on SpaceX for the First Time
Google's Project Suncatcher will launch TPU chips aboard SpaceX's Transporter-18 to test whether AI accelerators can survive and operate in orbit, exploring solar-powered space compute.
Why Gemini 3.5 Pro Is Stuck in Development Hell — and Why Google Is Betting Everything on Flash
Gemini 3.5 Pro is delayed, but Google is racing to iterate Flash. We analyze how Google is redefining the AI war — from benchmark competition to cost-per-task efficiency.
Deep Dives20 Years of Google Translate's Technical Evolution: From Trillion Tokens to TPUs to Gemini
Jeff Dean reflects on Google Translate's 20 years and three tech leaps: 2006's trillion-token language model validating Scaling Law, 2016's Seq2Seq+TPU neural translation, and now Gemini integration.