Benchmark

Benchmark methods

(1) CPU usage

CPU usage is calculated using process perspective measurement for the PID after launching App sense-stream. The result can exceed 100% for multi-threaded applications utilizing multiple cores simultaneously. CPU usage is sampled once per second within each minute, and the final value is the arithmetic mean of these samples.

  • Measurement: (process_cpu_time_delta / system_cpu_time_delta) × 100
  • Unit: %

(2) Memory usage

Memory usage is calculated by summing the PSS (Proportional Set Size) and swap memory of the PID after launching App sense-stream. PSS provides fair allocation of shared memory usage across processes. Memory usage is sampled once per second within each minute, and the final value is the arithmetic mean of these samples.

  • Measurement: PSS + SwapPss from /proc/pid/smaps
  • Unit: KB

(3) Inference time

Inference time refers to the time it takes to process a single window after launching App sense-stream. The measurement is log-based, parsing prediction_time values from application JSON logs. The number of measurements depends on the application’s push_period configuration. Inference time is measured over each minute, and the final value is the arithmetic mean of all measurements in that minute.

  • Measurement: JSON prediction_time field parsing from application stdout
  • Unit: ms

(4) Inference benchmark across devices

All benchmarks were conducted using the methods described above. The following table provides a consolidated overview of hardware specifications and benchmark results for the Cochl.Sense Edge SDK on each supported platform. Metrics include CPU configuration, memory, and average inference time based on the Cochl.Sense Edge SDK v1.6.0.

DeviceCPUMemoryCPU usage (%)Memory usage (KB)Inference time (ms)
Raspberry Pi 3ARM Cortex-A53 (4 cores, 1.2GHz)1 GB7.7954,797173.24
Raspberry Pi 4ARM Cortex-A72 (4 cores, 1.8GHz)4 GB6.7553,612156.48
NVIDIA Orin NanoARM v8 (6 cores, 1.5GHz)8 GB1.5257,35446.05
NVIDIA Jetson Orin NanoARM v8 (4 cores, 1.5GHz)4 GB4.4554,59394.34
Ubuntu 22 DesktopIntel Core i7-8700K (12 cores, 3.7GHz)64 GB0.0055,11813.26
Ubuntu 18 DesktopIntel Core i7-8700 (12 cores, 3.2GHz)32 GB0.4254,64810.40

Raspberry Pi 3 Model B Rev 1.2

  • CPU info: 4 Core 1.2GHz Cortex-A53, CoreMark score: 5597.39
  • RAM size: 1 GB
  • OS version: Debian GNU/Linux 11 (bullseye) (Debian Version: 11)
  • Kernel version: 6.1.21-v8+
  • System architecture: aarch64
  • Storage (Device model): SC32G
  • Ethernet interface info:
    • Interface: eth0
    • Speed: 100Mb/s
    • Driver: smsc95xx

CPU usage

rpi3_cpu_usage

Memory usage

rpi3_cpu_usage

Average inference time

rpi3_cpu_usage

Raspberry Pi 4 Model B Rev 1.4

  • CPU info: 4 Core 1.8GHz Cortex-A72, CoreMark score: 15371.61
  • RAM size: 4 GB
  • OS version: Debian GNU/Linux 11 (bullseye) (Debian Version: 11)
  • Kernel version: 6.1.21-v8+
  • System architecture: aarch64
  • Storage (Device model): SC32G
  • Ethernet interface info:
    • Interface: eth0
    • Speed: 1000Mb/s
    • Driver: bcmgenet

CPU usage

rpi4_cpu_usage

Memory usage

rpi4_cpu_usage

Average inference time

rpi4_cpu_usage

NVIDIA Orin Nano Developer Kit

  • CPU info: 6 Core 1.5GHz ARMv8 Processor rev 1 (v8l), CoreMark score: 23618.33
  • RAM size: 8 GB
  • OS version: Ubuntu 20.04.5 LTS (Debian Version: 20.04)
  • Kernel version: 5.10.104-tegra
  • System architecture: aarch64
  • Storage (Device model): FORESEE XP1000F256G
  • Ethernet interface info:
    • Interface: eth0
    • Speed: 1000Mb/s
    • Driver: r8169

CPU usage

nvidia_orin_cpu_usage

Memory usage

nvidia_orin_cpu_usage

Average inference time

nvidia_orin_cpu_usage

NVIDIA Jetson Orin Nano Developer Kit

  • CPU info: 4 Core 1.5GHz ARMv8 Processor rev 1 (v8l), CoreMark score: 10287.01
  • RAM size: 4 GB
  • OS version: Ubuntu 18.04.6 LTS (Debian Version: 18.04)
  • Kernel version: 4.9.201-tegra
  • System architecture: aarch64
  • Storage (Device model): SS32G
  • Ethernet interface info:
    • Interface: eth0
    • Speed: 1000Mb/s
    • Driver: r8168

CPU usage

nvidia_jetson_cpu_usage

Memory usage

nvidia_jetson_mem_usage

Average inference time

nvidia_jetson_inf_time

Linux-6.8.0-47-generic-x86_64-with-glibc2.35

  • CPU info: 12 Core 4.7GHz Intel(R) Core(TM) i7-8700K CPU @ 3.70GHz, CoreMark score: 52548.61
  • RAM size: 64 GB
  • OS version: Ubuntu 22.04.5 LTS (Debian Version: 22.04)
  • Kernel version: 6.8.0-47-generic
  • System architecture: x86_64
  • Storage (Device model): Samsung SSD 980 PRO 1TB
  • Ethernet interface info:
    • Interface: enp0s31f6
    • Speed: 1000Mb/s
    • Driver: e1000e

CPU usage

ubuntu22_cpu_usage

Memory usage

ubuntu22_cpu_usage

Average inference time

ubuntu22_cpu_usage

Linux-5.4.0-150-generic-x86_64-with-glibc2.27

  • CPU info: 12 Core 4.6GHz Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz, CoreMark score: 35207.14
  • RAM size: 32 GB
  • OS version: Ubuntu 18.04.6 LTS (Debian Version: 18.04)
  • Kernel version: 5.4.0-150-generic
  • System architecture: x86_64
  • Storage (Device model): Samsung SSD 860
  • Ethernet interface info:
    • Interface: enp2s0
    • Speed: 1000Mb/s
    • Driver: r8169

CPU usage

ubuntu18_cpu_usage

Memory usage

ubuntu18_cpu_usage

Average inference time

ubuntu18_cpu_usage