**Example case 1:** Running a benchmark pinned to a CPU core results in IPC=1.62: $ numactl -C 0 perf stat stress-ng --cpu 1 1.62 insn per cycle **Example case 2:** Running two instances of the benchmark pinned to two distinct CPU cores also results in IPC=1.62: $ numactl -C 0 perf stat stress-ng --cpu 1 & numactl -C 1 perf stat stress-ng --cpu 1; fg 1.62 insn per cycle 1.62 insn per cycle **Example case 3:** But running two instances of the benchmark pinned to the two distinct SMT threads provided by a single physical CPU core results in IPC=1.25: $ numactl -C 0 perf stat stress-ng --cpu 1 & numactl -C 8 perf stat stress-ng --cpu 1; fg 1.23 insn per cycle 1.25 insn per cycle **Base idea:** The cpugraph bars area can be used to paint the bars using a different color if suboptimal allocation of threads to logical CPUs is detected (case 3). **Suggested rendering for case 2:**  **Suggested rendering for case 3:**  For an overview of the SMT topic see for example http://www.anandtech.com/show/16261/investigating-performance-of-multithreading-on-zen-3-and-amd-ryzen-5000