Tencent / ncnn

ncnn is a high-performance neural network inference framework optimized for the mobile platform
Other
20.12k stars 4.14k forks source link

在两台rockchip rk3288上面测试benchmark,均达不到项目文档上面的速度,这是什么原因导致的 #3662

Open GloryKnight-K opened 2 years ago

GloryKnight-K commented 2 years ago
  1. rockchip rk3288上面实测benchmark速度 企业微信截图_16486350084633

企业微信截图_16486091605462

项目上面rockchip rk3288速度 企业微信截图_16485366557035

nihui commented 2 years ago

adb root adb shell stop

echo "performance" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor echo "performance" > /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor echo "performance" > /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor echo "performance" > /sys/devices/system/cpu/cpu3/cpufreq/scaling_governor

注意输入电压要足够

GloryKnight-K commented 2 years ago

adb root adb shell stop

echo "performance" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor echo "performance" > /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor echo "performance" > /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor echo "performance" > /sys/devices/system/cpu/cpu3/cpufreq/scaling_governor

注意输入电压要足够

电源是POE供电,调整了主频,速度有了较大上升,不过还是达不到文档上面速度,另外最高频实测sdk resnet18-in8模型速度要486ms,离benchmark速度差距较大,有什么办法可以缩小这差距吗 企业微信截图_16487933996552

nihui commented 2 years ago

top 看下是否有其他进程也在使用cpu资源 你的表格里4线程比2线程慢是不正常的