Running a little behind....

Just before I provide an update I wanted to pause for a few minutes and take the time to think out loud....

If I didn't work at a snail's pace on my T5810 project I'd have no chance of keeping up with my website write-ups.  I didn't want to focus solely on the T5810 project, or catching up with the articles either because I enjoy doing both.  That's what I thought, and as a result I feel as though I've fallen behind in both camps, the project and the website!  I did think about a journal-led approach for my website but when I thought about all the research that went into say choosing a new graphics card and the massive task of the ensuing write up, with so much detail that I'd probably miss something that was useful to you.  Instead, I think I should look back on what I've already done and reflect on the decisions made, like a sanity check and from that draw some conclusions.  That'll be helpful for me to plan better and hopefully make for a more interesting read.  This project will be more of what I did rather than what you should do, a journal rather than a tutorial.  Those are my thoughts for now but who knows?

Anyway, here's a little update on where I've got to with the T5810 project.  I'll probably end up writing detailed articles on the following:

1. Modified the case side panel so that it could house a modern-ish GPU.
2. NVMe upgrade and mirrored my daily driver OS for most efficient means of getting up and running because desktop settings and software were installed and tested ready to go anyway.
3. Purchased GPU power leads, which generally doesn't require any research but the T5810 did.
4. Cheeky cheap memory upgrade, multiple short tests including a fifty plus hour memory soak.  A fifty plus hour soak seemed a good way to test overall system stability and endurance (especially because I'd just repasted the CPU) at the same time as testing RAM.
5. A faulty machine and its replacement.
6. Installation and testing of the best GPU I had.  I've done a number of comparative timings which may be useful to people looking to buy a GPU for their own AI rig.
7. Research and purchase of a GPU which should be better than the one I have.  If it isn't, then we have a plan.  We optimise hardware and software whatever the outcome!
8. Research of replacement CPU has been completed today.

Whilst waiting for the GPU power leads to arrive I fixed my old (2011) daily driver machine, which the T5810 was originally intended to replace.  There were some good reasons for fixing a machine that I may not use for much longer which I'll go into later.

You've probably gathered from The News! article that the primary purpose of the Dell Precision Tower 5810 is for, basically local AI. 

Just as an aside I created a graphical link to my website from LinkedIn.  This graphic was created by AI, the main joke is that AIs have historically loved cats and it's a cat meme about AI.  Don't worry I won't be writing an article about the other little gags contained within the meme. 

It took a fair amount of prompting to alter the photo of Lucy, and it's not controversial to say, because it's largely understood now, that it probably took me as long in terms of prompting as what it would have taken me to make the alterations myself in a graphics package.  It didn't feel like that though because creating it in this way was fun, especially seeing what the AI came out with after each prompt.  I could've measured and compared that time to prove whether or not using AI really saved me time, but for a hobby that might be too analytical especially because I don't expect to be running a production line of bespoke cat memes anytime soon!  

If on the off-chance that I use AI again for my website I'll let you know that it's AI generated.  It's not appropriate for me to use AI for website content for quite a few reasons, the most obvious reason being I enjoy writing and I don't believe it would be appropriate to use AI written content on my website.  

The people I speak to in the tech community seem to either love AI or hate it.  I look at it in my "Windows-way".  That is to say, I prefer Linux to Windows as my daily-driver at home because it gets things done.  It just works.  It doesn't nanny me and there's nothing to work around, to work in the way I want to.  In other words it gets out of my way and works at my speed and helps.  Computing to me is turning on, waiting a few seconds and then being productive for the entire session.  All the computers I setup regardless of operating system work in that way.

Operating Systems are just tools and I will use Windows if that's the most efficient tool for a particular task.  In another article, just in case you don't already know, I'll show you how to create an image and verify it easily.  That image will be a Windows installation, we'll probably set up a dual boot system together step-by-step and we will be using Windows for a very specific task.

AI which is just another branch of technology, and technology in general, are just tools, it's not a case of someone being passionately right or wrong, or for using tech for tech's sake in a kind of vanity when it isn't appropriate.  The question is, does it do the job and does it do it well, well enough to do it better than what we have now.  At its most basic that philosophy is: Measure it, prove it, use it.

I have a number of future use cases for local AI and am sufficiently interested in AI to test it out for certain tasks as I have been doing since the days of Google's Bard.  I've proven that local AI (which is now only possible at all because, and thanks, to the much hard work and dedication of others) can be done fairly cheaply despite high NVMe and RAM and GPU costs, also because I don't really have any spare cash!  Cost savings are primarily what I want to show with these articles.  I expect that even second hand old hardware will jump in price imminently, even local AI on similar boxes will become impossible without spending a fair amount of money.  We can see this right now, take a look at the current price of T5820's on eBay.

My dream would be to run the latest cloud based frontier models on my Dell workstation, this may be possible, even right now.  But that dream is out of scope for the T5810 and surely wasn't considered when I designed the rig.  Or was it? 😀 We will see...