Speculative execution probably plays a large role in that. I think this is more of an average that you can expect to see. Additionally, if eager execution is n

Speculative execution probably plays a large role in that. I think this is more of an average that you can expect to see. Additionally, if eager execution is not at play, if a mispredict is triggered when data is finally fetched from main memory, the 5ns (handful of cycles) might just be the cost of unwinding, meaning the actual cost was 105ns.