-
Notifications
You must be signed in to change notification settings - Fork 19
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
The app is crashing when GPU acceleration is enabled using any Whisper model #137
Comments
Hi. Thanks for the report. Could you please run the app with
|
I applied the version change and it has fixed the crashes. However, when I record some audio and try to process it, the app is just processing it forever and not giving any text output. The processing animation in bottom left is going on and on but there is no output. If I check GPU usage with nvtop, as soon as I press 'Stop' after speaking the GPU usage spikes to 99% and is staying there forever. Pressing Cancel is also not stopping the loop. I have to close the app to get back the GPU to normal levels. This is happening with both Whisper Large and Whisper Tiny models. Attaching the log again for you to check:
|
There is a new version 4.5.0 released with newer ROCm... but don't install it! I haven't updated the Flatpak GPU add-on yet. I will let you know when it is ready for download. |
Ok, will that fix the issue I am facing? |
Maybe. This new version comes with updated ROCm libraries, so there is a chance that something will improve. Problem with the add-ons is resolved, so please update both |
OK, I have tried again after updating both the app and AMD addon. Still the same result unfortunately. The GPU usage goes to 99% and the app is stuck in the processing state for ages. The cancel button still does not work, the only way to stop the loop is to close the app. Attaching the log again:
|
Thanks for the log. It is weird a bit, but in the log every line is doubled. How it could happen? It looks like you started in the same time two instances of Speech Note 🤔 Apart from "doubled lines", it simply looks like your GPU is not well supported in ROCm. This is not too surprising, as AMD ROCm officially supports only a few graphics cards. Unfortunately, I don't think I can do anything more. |
I'm not sure why the logs are doubled, I just ran the app with
Maybe you can create an issue in the ROCm repository. Apart from that, nothing else to do so thanks for everything. |
It seems that requests for RX 5700XT support in ROCm were made repeatedly in the past and finally support was added.... but according to this, it was withdrawn starting with ROCm 5.3. There is also open ticket which says that in ROCm 6.2 support will be re-enabled. What a story :) |
Oh wow... that is such a lack of support from AMD. 5700 XT was the flagship model in its generation, and it has lost support! Unbelievable. Edit: Not sure if I should close the issue, since its not in your hands. Let me know. |
I'm somewhat surprised by the number of reported issues with AMD GPUs. I thought being Open Source guaranteed a little better outcome than NVidia, but it doesn't seem to be the case. |
Yeah... This could be AMD's strategy. They are very late in the field of GPU computing. Right now they are focusing ROCm only on a few graphics cards sold to data centers. They want to capture some of the money that goes to NVIDIA. Broad support for ROCm is not their priority. At least that's how I understand it. |
System Info
Distribution: Fedora Workstation 40
Desktop Environment: Gnome 46
GPU: AMD Radeon Rx 5700XT
CPU: AMD Ryzen 5 3600
RAM: 16 GB DDR4
Hi, just installed the app and it works flawlessly on CPU processing. However, when I enable GPU acceleration and use any of the Whisper models, the app freezes for 2 seconds and crashes after pressing 'Listen' button. I have installed the AMD addon flatpak and selected the ROC GPU acceleration method. When I disable GPU acceleration it works flawlessly again. Would appreciate any help. Thanks!
The text was updated successfully, but these errors were encountered: