Hi, my colleague and I both have Xavier NX’s with a custom image that are based on Jetpack 4.6.2. Our in house program deals with transcoding video and streaming it out, however I noticed even on the same code commit and power profile my Jetson was performing far worse.
Upon closer inspection I noticed that the NVENC and NVDEC on my device were running at ~500 and ~550Mhz respectively, whereas my colleages was running at ~730 and ~793Mhz. My VIC was also all over the place whereas theirs was static but I can pin that using the userspace control settings so that’s not so much of a concern.
My question is what may be causing my jetson to run so much slower? Is there any way to force it to run faster?
Hi,
By default NVDEC and NVENC are run in dynamic frequency scaling. You can enable maximum performance mode to run the engine fixed at maximum clock. If you use gstreamer, please enable the property:
[nvv4l2h264enc/nvv4l2h265enc]
maxperf-enable : Enable or Disable Max Performance mode
flags: readable, writable, changeable only in NULL or READY state
Boolean. Default: false
[nvv4l2decoder]
enable-max-performance: Set to enable max performance
flags: readable, writable
Boolean. Default: false
If you use jetson_multimedia_api, please check the similar options in 00 and 01 samples.
It would appear that I am already setting setMaxPerfMode for our encoder to 1. A bit more complex to set the decoder since we are using uridecodebin but I’m not sure it would have much effect.
I tried running the 01 sample to encode a test file with setMaxPerfMode(1) but the max frequency i saw was still only 499Mhz.
Okay, I tried my image on my sd card in my colleagues jetson and it ran fine and theirs in mine and I experienced the degredad performance. So it’s something wrong with my compute module itself. I will try and reflash to see if that fixes the issue, otherwise it appears to be some form of hardware degredation/failure…
Had a colleague run a full reflash via command line of 4.6.2 to reflash the QSPI, stuck my old sd card back in it and it seems to be fixed now.
A stray muon from outer space hitting the SPI memory and flipping a bit?
Anyway, case closed.