Meta cuts AI hardware costs by reusing server memory with custom ASIC

Meta is significantly reducing its hardware costs for AI inference workloads by repurposing memory from older servers. The company has developed a custom CXL ASIC that allows it to reuse this memory, leading to a 25% decrease in the number of machines required for certain tasks. AI

IMPACT This hardware efficiency improvement could lower the cost of deploying AI inference at scale.

RANK_REASON This is a hardware optimization and cost-saving measure for an AI company, not a core AI release or research.

Read on The Register — AI →

• CXL ASIC
• Mark Zuckerberg
• Meta

• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →