New GUI to monitor GPU health in linux mint

[mpdev-108]

Hello LinuxMint Development team,

I have come up with a mint native style GUI based GPU monitor to track the health of the GPU installed. Here's the code repo for review. I am a newbie here and I'd appreciate the detailed process for submitting the app for your review and subsequent inclusion in the mint Software Manager installs.

NOTE: There's quite a bit of vibe coding here. Let me know for any specific instructions with that.
Regards,
Madhu(he/him)

[Fantu]

Hi, you give the link to a long gist with features, how to install, use etc but i don't see one crucial thing, the link to the real software repository

[mpdev-108]

I will share it in the formal submission. I am here to know about the procedure to submit the app for review. I appreciate your help with that.

[mpdev-108]

I am a newbie to this community and I'd appreciate your guidance. My intention is to definitely share the code repo for your review by following the right process for the App inclusion. If this discussion itself is the starting point, then here's my repo with the code base for review.

https://github.com/mpdev-108/mintgpu-monitor/tree/feature/mvp_1

I'd appreciate your feedback/suggestions/pointers.

[Fantu]

Okay, now you've added the repository.
Unfortunately, with AI apparently allowing to do elaborate things without any prior knowledge, I'm seeing too many people continue to create new projects for existing things, but honestly, it seems like a waste of time with poor results. I've also seen similar projects ignored when they try to be added to Debian (Mint is based on Ubuntu, which is based on Debian, and software normally goes to Debian).

I've seen cases of projects with a person with no knowledge whatsoever who only works with AI. In one recent project, it even seemed good and quite used and would have been useful for my work, but then in practice that single "developer" (not a real developer) has no idea how to manage it, and even though it has perhaps hundreds of users, it's a software always in full development to which he continues to add things and has tons of bugs. Even though he recently made many fixes based on user reports, continuing to add and modify things has led to no stable enough version and it's not worth considering for serious use. And for the same thing there are many similar projects done with AI but none of them are good enough.

So basically if you want to make a contribution I think it would be better to first search in existing projects and maybe try to contribute to that, but without knowledge and only with AI you risk generating code that is not good enough and being more of a waste of time for those reviewing the contribution than a help.
It is not easy or quick to explain what would be best to do for those who have no knowledge.

[mpdev-108]

Hey, thanks for the detailed response—I appreciate the perspective.

I should've introduced myself initially. I've been a Backend software developer for the most part of my corporate career writing code mostly in java and to some extent in Python. So, I am fully aware of the full SDLC right from the requirements to production and beyond(post production support). Here's my linkedin profile in case you want to take a look https://www.linkedin.com/in/madhu-kiran-442ba315.

I’m not relying on AI alone, and I fully agree with your point about long-term maintainability.

My goal with this project was fairly specific: I’ve been using Linux Mint for 4–5 years, and after setting up an NVIDIA GPU for running local LLMs (via Ollama), I noticed the lack of a lightweight GUI tool for monitoring GPU usage—something similar to a task manager view. That’s what led me to explore building this.

That said, I do understand your point about contributing to existing projects first. If there are established tools or Mint/Debian projects in this space that you’d recommend contributing to, I’d be happy to explore those.

Also, from your experience, what are the key expectations or areas of knowledge for contributing to Mint/Debian projects?

I'll go through https://linuxmint-developer-guide.readthedocs.io/en/latest/ to see what I can contribute to.

I appreciate your time and guidance.

Cheers,
Madhu(He/Him)

[Fantu]

So if you already have programming experience, it's an excellent starting point for guiding and correcting AI. The first essential thing is that LLMs, even if they're called "AI," they are not real artificial intelligence at all and, fundamentally, they never will be, even if we greatly increase the resources they use and the amount of data they have access to and on which they have been trained.

I've been using them for a while, but perhaps I've understood the basics enough to try to use them to achieve some decent results. However, I also have to be careful, because errors can always arise, even when they appear to produce excellent results.

First of all, to have a good chance of producing correct results, it must have correct data available and be able to test what it produces, even making multiple attempts until it produces something at least working.
Therefore, I have already seen a huge difference by investing time in preparing the local repositories of the projects it has to work on, instructions on some essential things, repositories also of the dependencies, libraries and frameworks used, and having it practically check there the functions it uses to ensure they are existing and correct. Otherwise, it easily relies on outdated, incomplete or incorrect information, even searching on the Internet (in many cases, by searching it has a greater chance of making mistakes because it cannot understand whether any sources are correct and good).
It is always necessary to provide it with test environments on which he can run tests, often better if specific, it is also better not to have him test on the host but to use containers, VMs or something else dedicated, clean and isolated.

For example, yesterday I tried to use fable 5 for some fixes and additions to a nextcloud app that I needed for work, it seems to me that I achieved quite good results but I saw that it was important to prepare everything beforehand (over 2 hours even just the initial preparation before starting) and I saw that it had many problems both in the builds and in the tests and in some patches that had to be modified during, so without that part I would have wasted a lot of time and all the tokens without making it do anything (like one of the first things I tried months ago in which in 4 hours I was unable to make it do, continuing to make it try, not even a correct script that perhaps without AI in the same way I would have done on my own with manual searches).
So in result I probably reached the main things I needed in 5 hours, without I suppose it would have taken me 20 hours to do the same, I worked a lot of hours on php but many years ago (at the time of php 4 and 5) and in the last few years almost only some backport fixes.

Once you know how to have a good chance of producing something decent, you have to choose what's best to do.
I'd recommend something you have experience with rather than something you don't, and something you already use and will use.
Because beyond letting the AI ​​produce, you need to monitor and test what it produce and the contributions you intend to make using AI to avoid wasting developers' time.

Another important thing is to try to limit what produce and also do targeted things. If you do a lot of big things because it seems like a lot can be done quickly and easily, you risk lowering the quality and making it difficult to manage effectively, whether it's a whole project from scratch or a contribution to an existing project.

More specifically, AI tends to lose context the more the data expands and basically to reduce both consumption and targeted results it is divided into many small parts and basically it is good for less token consumption and in most cases good results but it must also be taken into account that in this way it also has a "narrow" view of the total and risks producing duplicated, excessive, conflicting things with other parts etc... it is a bit complicated to explain everything though.

Mint has done a lot in Python and you can help with many things, there is need of help from other contributors but you also have to be careful not to waste the Mint team's time because it is small and already too busy to waste time on "AI slop".

Regarding this program you are interested in, there are several projects, some specific for hw (amd or nvidia) and others generic.
For example a generic one that doesn't seem bad to me from a quick look (I not tried it) is https://missioncenter.io/ https://gitlab.com/mission-center-devs/mission-center someone also tried to add it in Debian but need to add some missing dependencies first: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1107258

EDIT:
One of the things about "AI" that I hate the most is how bad it has made all the translators (since they switched to using LLMs as a basis), it modifies the sentences, cuts, changes the meaning, it's stressful, I don't know if I managed to write correctly and comprehensively what I wanted, I had to make several manual corrections but in several sentences pieces were lost and it seems not clear enough to me

[mpdev-108]

Hey, thanks for taking the time to share all this detail and for the pointers (especially Mission Center and the Debian packaging context). I really appreciate the explanation.

I'm still learning how to work effectively with AI in real development, and it took me several iterations to reach the initial version I submitted. I'm fine if this submission isn't accepted—that's part of the process.

I'll go through the pointers you provided and see what I can contribute. Either way, thank you for the time and effort you put into explaining this.

Cheers,