I regret building a $3000 Pi AI Cluster

It's been a while. Some Kickstarters take time.

What do you think about doing the same thing 10 times in a row, over and over? If that's your jam, build a cluster like I did in this video!

If you want to replicate the Compute Blade Cluster, here are all the parts I used (some links are affiliate links):

- Compute Blade DEV: https://www.pishop.us/product/compute-blade-dev/
- Compute Blade Standard Fan Unit: https://www.pishop.us/product/fansta-1v1-0/
- Compute Blade 10" 3D Print Rackmount: https://github.com/Uptime-Lab/compute-blade/tree/main/models/bladerunner
- Raspberry Pi CM5 16GB (CM5016000): https://www.pishop.us/product/raspberry-pi-compute-module-5-16gb-ram-lite-cm5016000/
- GLOTRENDS Aluminum CM5 Heatsink: https://amzn.to/45YACBV
- Patriot P300 256GB NVMe SSD 10-pack: https://amzn.to/4lPJ0JC
- GigaPlus 2.5 Gbps 10 port PoE+ switch: https://amzn.to/3UNOwSd
- GigaPlus 10" Rack Mount 3D Print ears: https://www.printables.com/model/1215585-unified-10-rack-gigaplus-switch-mounting-ears
- Monoprice Cat6A SlimRun 6" Cat6 patch cables (10 pack): https://amzn.to/4fX1gzr
- ioplex SFP+ Twinax DAC patch cable: https://amzn.to/47UlPKX
- DeskPi RackMate TT: https://amzn.to/3UUjCHP

Other resources mentioned in the video:

- My older Compute Blade video: https://www.youtube.com/watch?v=rKDGlpnP-vE
- Xerxes Pi Kickstarter: https://www.kickstarter.com/projects/1907647187/small-board-big-possibilities-xerxes-pi
- Xerxes Pi testing: https://github.com/geerlingguy/raspberry-pi-pcie-devices/issues/753
- Unredacted Labs' Tor exit cluster: https://unredacted.org/blog/2025/05/unredacted-labs/
- World's Largest Pi Cluster at UC Santa Barbara: https://www.independent.com/2025/04/29/worlds-biggest-raspberry-pi-cluster-is-now-at-uc-santa-barbara/

Support me on Patreon: https://www.patreon.com/geerlingguy
Sponsor me on GitHub: https://github.com/sponsors/geerlingguy
Merch: https://www.redshirtjeff.com
2nd Channel: https://www.youtube.com/@GeerlingEngineering
3rd Channel: https://www.youtube.com/@Level2Jeff

Contents:

00:00 - 2 years 2 long
00:32 - Another blade
01:30 - Unboxing a cluster
03:06 - Assembling one of 10
06:11 - A bandolier of microSD cards
07:16 - Assembling blades 2-10
08:16 - Next steps - fans and storage
10:17 - Flashing NVMe drives
11:00 - Only Fan Units
12:45 - They all just... disappeared
14:33 - Spending another $200
15:45 - Cluster benchmarks: HPL
16:23 - Spending another $70
17:03 - Framework cluster vs Pi cluster
18:12 - AI clustering (down one node)
19:26 - llama.cpp is totally RPC
20:54 - Exo and Distributed Llama
22:09 - Blade (the movie) vs blade (the compute)