search

Qengineering / Jover

Overclocking the Jetson Nano CPU and GPU
https://qengineering.eu/overclocking-the-jetson-nano.html
BSD 3-Clause "New" or "Revised" License
19 stars 2 forks source link

do i have to use an available freq? #3

Open xrstokes opened 2 years ago

xrstokes commented 2 years ago

/sys/devices/57000000.gpu/devfreq/57000000.gpu/available_frequencies 76800000 153600000 230400000 307200000 384000000 460800000 537600000 614400000 691200000 768000000 844800000 921600000 998400000

I don't seem to have any errors at 998.4 but you say in the post 950. if i select 921.6 i'm not over stock freq though. Can i add another in the middle?

Thanks so much for your work and effort that went into this.

Qengineering commented 2 years ago

It's all a matter of trail and error. The max frequency will work most of the time. However, when running some deep learning model with 100% CPU and 100% GPU (CUDA) we experienced some issues at 998.4 MHz. Hence the 950 in the post. No need to put a new value in the middle. You can give whatever frequencies you like echo 950000000 > /sys/devices/57000000.gpu/devfreq/57000000.gpu/max_freq in the file overclock.sh.

xrstokes commented 2 years ago

when i tried to set it to 950 it seems to jump to 998. same with cpu. it jumps to the closest available. in my case 1912Mhz

Package: nvidia-jetpack Version: 4.5-b129 Architecture: arm64 Maintainer: NVIDIA Corporation Installed-Size: 194 Depends: nvidia-cuda (= 4.5-b129), nvidia-opencv (= 4.5-b129), nvidia-cudnn8 (= 4.5-b129), nvidia-tensorrt (= 4.5-b129), nvidia-visionworks (= 4.5-b129), nvidia-container (= 4.5-b129), nvidia-vpi (= 4.5-b129), nvidia-l4t-jetson-multimedia-api (>> 32.5-0), nvidia-l4t-jetson-multimedia-api (<< 32.6-0) Homepage: http://developer.nvidia.com/jetson Priority: standard Section: metapackages Filename: pool/main/n/nvidia-jetpack/nvidia-jetpack_4.5-b129_arm64.deb Size: 29358 SHA256: 9ee354a66d932a3fbb244c926f333143a845c627c6981d108e01df2958ac462c SHA1: 0e07f27c6fb9e34a70c69ae1150d1e578e938089 MD5sum: a551bbc8ff653c8983ce1804082bbcab Description: NVIDIA Jetpack Meta Package Description-md5: ad1462289bdbc54909ae109d1d32c0a8

Qengineering commented 2 years ago

OK, seems NVIDIA has changed the governor since i wrote the post. Leaves you with two options, both I didn't test.

  1. Alter the 998400000 to a lower number.
  2. Place a extra value in the sequence.

Both methods can cause the PLL to malfunction. Or just accept the 921 MHz, and drop the extra 3%