not yet;
based on what I see in the past (i.e., issues mixing bullseye and bookworm containers), I prefer wating for Chromium:3 bookworm before trying again.
Hi @henrique.tx
yes I tried.
Basically the Chromium Bookworm Container works for a simple web page.
But the behavior with 3D content is not ok.
It seems that Vulkan is not enabled (or something like that), but I need some more time to investigate.
I let you know.
Hi @jeremias.tx
sorry for the confusion.
This topic has been created when I’ve been working on the Mini (February 2023).
But in the meanwhile I switched to the Plus (tha has HDMI).
For this reason now I refer to the Plus (that has Vulkan).
As we saw together with @matheus.tx it seems that even if chromium flags show that WebGL and WebGL2 are enabled (and Vulkan is not)
running this simple 3D js application Basic demo - CodeSandbox show a high load on the CPU.
This is unexpected, but it’s difficult to say since at the moment a GPU profiling is not available on TorizonCore for Verdin IMX8M-Plus.
@jeremias.tx Are you able to reproduce this behavior (high load on CPU when running the example) on your side?
Is it possible that WebGL is enabled, but not used (for some reason)?
Does the demo on that webpage actually run for you? I tried myself and the webpage opens but the demo actually never seems to start. In some cases after a while the webpage just goes black for me.
Yes, it works on the Plus, with one lof the nightlies of TorizonCore downloaded more or less on the middle of July 2023 (I can’t check now the exat number).
I was able to get the demo application to run. Though I had to restart the container multiple times as sometimes the demo would fail to load or the webpage would freeze/crash. Is this what you’ve observed as well or does it work for you every time?
On the matter of CPU usage here’s what I see while the demo app is running. According to top I see 3 chromium processes at the top of the list that average around 70%, 50%, and 25% cpu usage. Though occasionally I see these values spike higher for short periods. Is this what you are seeing in terms of “high cpu load”?
During my test, the demo works well. No freeze/crash.
For the cpu load, I confirm what you see.
And it’s not expected, since the whole animation of the demo is supposed to be done by the GPU, and so the CPU should stay more or less “unloaded”.
As a suggestion from muy side, a GPU profiling tool in TorizonCore would simplify dramatically this kind of analysis.
My feeling (it’s only a feeling) is that the animation is done by the CPU, even if the GPU is enabled.
Is there a way to confirm or negate this?
If it’s true, why the GPU is not used?
I did some additional testing and it really seems like something here is amiss with regards to hardware acceleration. I used this site to test WebGL: WebGL Blobs
With this I see a similarly high cpu usage as well. I’ll report this to our development team to look into.
As a suggestion from muy side, a GPU profiling tool in TorizonCore would simplify dramatically this kind of analysis.
It seems that this tool could be included in TorizonCore, but I think that now it’s not.
Searching in NXP Community I found this topic and I read
gputop and libgpuperfcnt builds for linux BSP seem broken in 5.15 indeed
I opened a ticket to NXP to investigate what happens to gputop.
NXP Tech support wrote me that
current bsp supports gputop already, I tested 5.15 without any issue, you can test, if you have any issue, pls tell me what you get, let me reproduce this
@jeremias.tx do you know why gputop has not been included in TorizonCore for iMX8M-Plus?
do you know why gputop has not been included in TorizonCore for iMX8M-Plus?
Well there’s a multitude of reasons.
The tool itself is somewhat unreliable and prone to issues as you’ve seen on the NXP forums.
Toradex can’t reasonably support, and maintain this tool if there are issues or questions since it’s not in our control or design.
It’s not a tool you’d want in a production ready image like TorizonCore so having it there by default is not desirable for a lot of customers.
That said there’s nothing stopping you from including this yourself. You could even try building this and deploying the tool in a container. In fact there were some experiments to try this by some in Toradex: Docker
You could give this a try but again it’s not an officially supported tool from us and the support we can provide is minimal at best. Any advanced issues or questions regarding the tool or GPU in general would best be brought to NXP as the proprietary nature here limits what we can do.
Nothing significant to share at the moment. The team is fairly sure there is something strange going on with hardware acceleration in chromium, though there’s no exact details and further work and investigation is still needed.
).