EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Lobsters·Aug 26·5 min read
Lobsters·Aug 26·5 min readMemory ordering in CPUs
TL;DR
Memory ordering in CPUs varies significantly between architectures, with strong ordering (like x86) and weak ordering (like ARM) leading to different scalability assumptions. Many CPUs do not strictly adhere to their memory models, often optimizing for performance instead.
✦ Why It Matters
Evaluate your code's assumptions about memory ordering to avoid concurrency issues.
Key Takeaways
- Understand the difference between strong and weak memory models.
- Optimize for reduced contention in multi-threaded applications.
- Recognize that both architectures may not strictly follow their memory models.
- Evaluate performance based on real workloads rather than theoretical advantages.
How It Works
CPUs often execute memory operations optimistically, assuming most accesses are uncontended. They track potential ordering violations with metadata, allowing for reordering of operations unless contention is detected.