Google porting all internal workloads to Arm, with help from GenAI (www.theregister.com)

🤖 AI Summary
Google has revealed it has ported about 30,000 production packages to Arm and plans to migrate the rest of its internal workloads so they can run on both x86 and its homegrown Axion Arm CPUs—YouTube, Gmail and BigQuery already run on both architectures. In a preprint titled “Instruction Set Migration at Warehouse Scale” and a blog post, engineers say early fears about massive architecture-specific bugs (floating-point drift, concurrency, intrinsics) were largely mitigated by modern compilers and sanitizers. The bulk of work instead involved fixing brittle tests, updating long-tail build and release systems, stabilizing rollouts, and avoiding disruption to critical services. Google’s goal is to let its Borg cluster manager (the antecedent to Kubernetes) schedule across Arm and x86 to maximize efficiency. To scale the effort across ~100,000 packages, Google repurposed automation and built an AI assistant called CogniPort that inspects build and test failures and proposes migration commits. CogniPort succeeded roughly 30% of the time on tasks like test fixes, platform-specific conditionals, and data representation issues. The shift matters because Axion machines reportedly provide up to 65% better price-performance and ~60% better energy efficiency than x86, implying significant cost and power savings and likely reduced dependency on x86 servers—while highlighting how compiler advances and AI-driven developer tooling can accelerate massive ISA migrations.
Loading comments...
loading comments...