I don't think you should be an expert in CPUs architectures but but having an understanding of memory layout and how instruction are dispatched is not very hard or time consuming.
I see many people having trouble to understand the difference between value and reference, what pointers are, why it's better for structures to contain variables of the same type and have a certain size, why is better to call a function once for a large chunk of data instead of calling it many times for small chunks of data, why iterative or tail call recursivity are to be preffered over simple recursive functions.
The view is most LOB apps won't care about performance because they are waiting for i/o so we should not care about performance but coding speed, OOP, Uncle Bob's principles, GoF patterns, Restful and trendy architecture of the day. While I sure that coding speed matters, a sound architecture matters, I also think that throughput matters, that minimizing delays matters and that some problems are better dealed with by thinking of the data and how the CPUs likes to access data and work with it instead of just firing up more Kubernetes pods hoping that scaling will get rid of performance issues. By not thinking about the data and the CPU we didn't get rid of the complexity we just moved it to the infrastructure and in code having to deal with a more complex infrastructure.
I see many people having trouble to understand the difference between value and reference, what pointers are, why it's better for structures to contain variables of the same type and have a certain size, why is better to call a function once for a large chunk of data instead of calling it many times for small chunks of data, why iterative or tail call recursivity are to be preffered over simple recursive functions.
The view is most LOB apps won't care about performance because they are waiting for i/o so we should not care about performance but coding speed, OOP, Uncle Bob's principles, GoF patterns, Restful and trendy architecture of the day. While I sure that coding speed matters, a sound architecture matters, I also think that throughput matters, that minimizing delays matters and that some problems are better dealed with by thinking of the data and how the CPUs likes to access data and work with it instead of just firing up more Kubernetes pods hoping that scaling will get rid of performance issues. By not thinking about the data and the CPU we didn't get rid of the complexity we just moved it to the infrastructure and in code having to deal with a more complex infrastructure.