If you trust this link, click it to continue.
https://semiengineering.com/llm-inference-core-bottlenecks-imposed-by-memory-compute-capacity-synchronization-overheads-nvidia/
