Understanding your Processor (1/n)
summary
A microbenchmark reads array elements once in random order and accumulates them, pinned to a single hardware thread, and plots bytes/sec versus array size. Results show throughput near 20 GB/s for data that fits in L1 cache then collapsing to about 450 MB/s once accesses miss through L2/L3 and hit main memory. The demo uses simple read/add loops and a chart; it reliably reproduces cache-size cliffs and recommends keeping hot working sets small to avoid evicting other hot data.
AI-written summary · about
related talks
01h 30mTuning C++: Benchmarks, and CPUs, and Compilers! Oh My!
54mx86 Internals for Fun & Profit
01h 04mDesign for Performance
47mOptimizing Multithreading Performance
05mMaximising Throughput & Minimising Latency in High-Performance C++ Code
35mAlgorithmic and microarchitecture optimizations of C++ applications
49mAlgorithmic and microarchitecture optimizations of C++ applications
01h 37mAn Overview of Program Optimization Techniques