Quickly measure CPU throughput and latency with a repeatable command-line test. 16.02.2026 | reading time: 2 min Want a fast, repeatable CPU stress and throughput check? Sysbench is a small command-line tool that runs a configurable CPU workload and reports events per second and latency; run it to compare kernels, VM types or tuning changes. Hands-on example Run this on a test system to measure a 4-thread workload: ``` sysbench cpu --threads=4 --cpu-max-prime=20000 run ```A representative run produces output like: ``` sysbench 1.0.21 (using LuaJIT 2.0.4) Running the test with following options: Number of threads: 4 Prime numbers limit: 20000 Threads started! CPU speed: events per second: 102.18 General statistics: total time: 10.0000s total number of events: 1022 Latency (ms): min: 0.54 avg: 3.90 max: 20.32 95th percentile: 10.41 Threads fairness: events (avg/stddev): 255.5000/1.50 execution time (avg/stddev): 2.5550/0.01 ``` Tune the test Change `--threads` to test parallelism, use `--cpu-max-prime` to change computation intensity, or set `--time` to run longer; pin workers with `taskset` for reproducible single-socket or core measurements, repeat runs and record events-per-second and latency to spot regressions or scheduler effects. When sysbench isn't enough Sysbench is great for synthetic integer workloads, but real applications differ; complement it with sampling and tracing tools to find bottlenecks, and avoid over-interpreting raw events-per-second as application throughput without workload context. Wrap-up Run a few controlled sysbench tests, tune parameters, then compare results across kernels, VM types or CPU governors to draw conclusions; keep measuring and learning, and consider studying for CompTIA Linux+ or LPIC-1 with intensive exam prep at bitsandbytes.academy to deepen your Linux skills. Join Bits & Bytes Academy First class LINUX exam preparation. utilities processes infrastructure