I’m just saying. She could have totally stole them from Mexicans.

Which would technically not be a lie.

I usually follow the craft computing video whenever I have to set it up. He has a document in the description with all the things you might need to passthrough a GPU but only like half are needed to passthrough an HBA

https://www.youtube.com/watch?v=_hOBAGKLQkI

Proxmox wiki has a more concise guide IMO

https://pve.proxmox.com/wiki/PCI_Passthrough

It was really simple to do in Proxmox.

You will find no name brand HBAs in IT mode on eBay for half the price of Intel, Supermicro, Dell, Etc branded ones. Do not buy the no names. I spent a week flashing and reflashing some cheap one, cycling through cables, etc. Nothing.

My supermicro branded one worked absolutely no issue. And I think it was like $40

It probably took a total of 30 minutes to pass it through and build the VM and everything. It took a couple days to rebuild my data from my previous truenas server but I had 10 TB of data on 4 drives.

The only issues I’ve had have been my own reading comprehension in setting up truenas accounts.

Are you using truenas as the entire homelab?

I also love messing with stuff until it breaks and I learn something, but I’ve decided I just want my files to be accessible.

So I actually have truenas virtualized with a passed through HBA so I can run proxmox to host all my breakable VMs while leaving truenas alone.

Completely fair

“I’d rather my kids be autistic than dead”

I think this is a completely valid train of thought.

Misguided for sure, but valid.

From what I understand its not as fast as a consumer Nvdia card but but close.

And you can have much more “Vram” because they do unified memory. I think the max is 75% of total system memory goes to the GPU. So a top spec Mac mini M4 Pro with 48GB of Ram would have 32gb dedicated to GPU/NPU tasks for $2000

Compare that to JUST a 5090 32GB for $2000 MSRP and its pretty compelling.

$200 and its the 64GB model with 2x 4090’s amounts of Vram.

Its certainly better than the AMD AI experience and its the best price for getting into AI stuff so says nerds with more money and experience than me.

Honestly if you’re not gaming or playing with new hardware, there is absolutely no point.

I’ve considered swapping this computer over to Fedora for a hot minute, but it really is a gaming PC and I should stop trying to break it.

True, but I have an addiction and that’s buying stuff to cope with all the drawbacks of late stage capitalism.

I am but a consumer who must be given reasons to consume.

The Lenovo Thinkcentre M715q were $400 total after upgrades. I fortunately had 3 32 GB kits of ram from my work’s e-waste bin but if I had to add those it would probably be $550 ish The rack was $120 from 52pi I bought 2 extra 10in shelves for $25 each the Pi cluster rack was also $50 (shit I thought it was $20. Not worth) Patch Panel was $20 There’s a UPS that was $80 And the switch was $80

So in total I spent $800 on this set up

To fully replicate from scratch you would need to spend $160 on raspberry pis and probably $20 on cables

So $1000 theoratically

Gonna write my short story about the orc barbarians who destroy human colonies that get too close to orc territory, not because they’re inherently evil, but because they’ve seen what human greed for power and domination does to subjugated races, the flow of magic, and the health of the earth. So they view humans as evil.

“Your kind knows nothing but exploitation! You drain the lands of their nutrients to feed cities of sycophants until they are fat! Tell me, adventurer, when was the last time you heard of a dragon attacking an orc caravan? We have no fear of such beings as they only attack the depraved greed of man.”

“Attacked the village? Do your handlers even lie to hired blades? Yes we burned the village you call Argath, but no one was harmed. Humans, as dangerous as you are, are still cowards. Surrounding a mining village and telling them to leave when they’re outnumbered ten to one is hardly, what you would call, a negotiation. We sent hunters to escort them out of the mountains of Gri’ut Kar and burned the village to ensure the trek was one way.”

The PIs were honestly because I had them.

I think I’d rather use them for something else like robotics or a Birdnet pi.

But the pi rack was like $20 and hilarious.

The objectively correct answer for more compute is more mini PCs though. And I’m really thinking about the Mac Mini option for AI.

Ollama and all that runs on it its just the firewall rules and opening it up to my network that’s the issue.

I cannot get ufw, iptables, or anything like that running on it. So I usually just ssh into the PC and do a CLI only interaction. Which is mostly fine.

I want to use OpenWebUI so I can feed it notes and books as context, but I need the API which isn’t open on my network.

I was thinking about that now that I have Mac Minis on the mind. I might even just set a mac mini on top next to the modem.

Ollama + Gemma/Deepseek is a great start. I have only ran AI on my AMD 6600XT and that wasn’t great and everything that I know is that AMD is fine for gaming AI tasks these days and not really LLM or Gen AI tasks.

A RTX 3060 12gb is the easiest and best self hosted option in my opinion. New for <$300 and used even less. However, I was running with a Geforce 1660 ti for a while and thats <$100

A mac is a very funny and objectively correct option

I think I’m going to have a harder time fitting a threadripper in my 10 inch rack than I am getting any GPU in there.

I do already have a NAS. It’s in another box in my office.

I was considering replacing the PIs with a BOD and passing that through to one of my boxes via USB and virtualizing something. I compromised by putting 2tb Sata SSDs in each box to use for database stuff and then backing that up to the spinning rust in the other room.

How do I do that? Good question. I take suggestions.

With a RTX 3060 12gb, I have been perfectly happy with the quality and speed of the responses. It’s much slower than my 5060ti which I think is the sweet spot for text based LLM tasks. A larger context window provided by more vram or a web based AI is cool and useful, but I haven’t found the need to do that yet in my use case.

As you may have guessed, I can’t fit a 3060 in this rack. That’s in a different server that houses my NAS. I have done AI on my 2018 Epyc server CPU and its just not usable. Even with 109gb of ram, not usable. Even clustered, I wouldn’t try running anything on these machines. They are for docker containers and minecraft servers. Jeff Geerling probably has a video on trying to run an AI on a bunch of Raspberry Pis. I just saw his video using Ryzen AI Strix boards and that was ass compared to my 3060.

But to my use case, I am just asking AI to generate simple scripts based on manuals I feed it or some sort of writing task. I either get it to take my notes on a topic and make an outline that makes sense and I fill it in or I feed it finished writings and ask for grammatical or tone fixes. Thats fucking it and it boggles my mind that anyone is doing anything more intensive then that. I am not training anything and 12gb VRAM is plenty if I wanna feed like 10-100 pages of context. Would it be better with a 4090? Probably, but for my uses I haven’t noticed a difference in quality between my local LLM and the web based stuff.

That’s fair and justified. I have the label maker right now in my hands. I can fix this at any moment and yet I choose not to.

I’m man feeding orphans to the orphan crushing machine. I can stop this at any moment.