Inmy_aimlbySanjay Basu, PhD·1d agoPerformance per Whose Dollar?TPUs, agentic inference, and the end of the one-chip cloud
Harshit Anand·Oct 21PyTorch now runs natively on TPUs, and it’s open sourceGoogle and Meta’s PyTorch team open-sourced TorchTPU. Here’s what ships, what the numbers say, and where I’d stay careful if you build or…
xbill·Sep 26Google’s QAT Gemma 4 26B-A4B on One TPU v6e: 15.6x the KV Cache and 1.9x the Throughput of FP8This article provides a step by step guide to serving Google’s quantization-aware-trained (QAT) Gemma 4 26B-A4B on one Google Cloud TPU v6e…
InGenerative AIbyxbill·Sep 24Running a Jev-Style Decision Model on One TPU v6e: What Fits, What It Costs, and What Changes From…This article provides a step-by-step guide to running a Jev-style decision model on one Google Cloud TPU v6e chip with Gemma 4 and vLLM…
Ayaka·Aug 25Removing the redundant Pallas TPU v4 Megacore entry synchronizationPallas adds an extra synchronization sequence
