This was found being given away on FB Marketplace as it "didn't work"
The case is huge and not the best looking, but does have blue LED's in the fans
It has a clear perspex window in one side which falls out all the time
USB, power buttons and audio on the top
The previous owner had taken out the GPU and put in an old power supply but didn't screw it in to the case
The mother board before and after the cables were moved out the way
Intergrator 500W power supply
Anyway I powered this up and yes, nothing happened apart from the fans starting up. I reseated the RAM and reset the BIOS and it fired straight up, turns out it's a seventh gen i7 running at 3.6GHz with 16GB of RAM, and I put an old GPU in it. I'm tempted to recase the motherboard and use this to replace the PC running my NVR
Update
I decided to go in a different direction and start running a local AI model with this PC, it will only be used for image analysis from my security cameras in Home Assistant
As you are no doubt aware to use an AI model on a PC a decent GPU is needed, So me being me and not wanting to spend big money I decided to try the old AMD Radeon HD 7700 Cape Verde Series GPU I got with the other Gigabyte PC that was donated to the museum a while ago just to see if the idea was feasible or not
If you try to research the idea online you'll see comments that say you need at least a £2000 GPU and the latest generation CPU with 32 GB RAM before you can even think about using any kind of AI model at home... So lets try with a GPU I got free off FB Marketplace that was originally in an second gen i3 parts special PC
I installed the latest Ubuntu Server as this will be a headless machine, then installed Ollama, and then tried a few AI models that fit in the tiny 1GB VRAM the card has and the best seems to be minicpm-v4.6:1b, as I only need a short description of what's in the image, it'll either be nothing, a human, or a cat. I had tried Qwen, LLaMA, and Gemma models as well and they either take too long thinking or ramble on for hours about what it's expecting to see but can't
Using the above GPU and AI model it would analyse the image and provide a text description in about ten seconds, it's a little slow but workable
So you don't need massive power to analyse security camera images.
Next I bought a slightly better GPU from Vinted for £20. I'll admit I know nothing about modern GPU's so just picked one that had more VRAM and more of everything else specs wise. It's an Asus branded AMD Radeon R9 280X GPU with 3GB VRAM
And then I came across the first problem, this GPU needs two plugs from the PSU, and more power which the Intergrator 500W couldn't supply as it only had the six pin plug available (with the detachable two pin bit)
Again I know nothing about modern PC power supplies, so took as many photos as I thought I would need then visited the local independant computer shop who recommended this power supply which cost £60. I know I could have got it cheaper online but I want it NOW!
850W of power and a load of extra plugs I won't need at the moment
Anyway after all that expense I now get a text description of the photo in about six seconds
Recap of hardware
Gigabyte H110M motherboard
Seventh Gen i7-7700 CPU running at 3.60GHz with 4.2GHz peak
16GB DDR4 Ram
Asus branded AMD Radeon R9 280X GPU with 3GB VRAM
250GB SSD with the latest Ubuntu Server installed
Ollama software to install the AI model which is minicpm-v4.6:1b, others are available
New power supply that can power everything
Weird looking case which is too big and cumbersome
Cost to the museum so far is £80 ish and most of that was for the power supply
Notes
The 1b at the end of the AI model name means it will fit in 1GB VRAM, if you have more VRAM you can go up here, the AI model really needs to fit completely in VRAM or it will be tragically slow, so if you have 8GB VRAM you can run an 8b model
The first time you send an image to be analysed it adds about ten seconds to load the model which will stay in VRAM for the next time
You'll need the Ollama integration in Home Assistant, and you need to set up the AI Task service which is what you use to analyse the image
You will not need to set up the Ollama Conversation service, you wouldn't get much sense from a 1b model anyway, but it's fine for image analysis
Now I just need to rewrite all my Home Assistant scripts to use the local AI rather than Qwen through Groq, and I still have another i7 PC with a spare GPU and 500W power supply to find a use for..?
O - Invalid stream, 110:1
Welcome to the comment section for Gigabyteh110m.
Please note: I am unable to reply to your comments here, so if you have any questions about Gigabyteh110m, then please contact me using the feedback option in the share menu below
No comments yet, be the first to start the conversation for Gigabyteh110m