Server Build for Education/Profit

Hello Everyone,

I am writing in for a little help with building a network server. I am currently in a Full Stack Web Development student and I am trying to hack together a small server with parts left over from a GPU based crypto currency mine. The purpose of this server is both educational and practical. This will be a small load server that will hold my personal projects, personal and small business website, and needs to be scaleable to test multi user media applications and act as a personal repository. I have the following equipment available.

1x Full size + 1 half size server case.
6x 4u empty Rosewell server cases setup to hold eight GPU’s each.

6x EVGA 1600 watt power supplies

1x Cisco ASA Router

1x Ubiquity Unifi Security Gateway Pro

1x Cisco SG300-28

1x Cisco SG200-26P

A Plethora of HDD’s/SSD’s

Again, I would like to build the system to be, as best as possible, upgradeable. I believe almost everything is here except a UPS and the actual server build. What would you all recommend for a UPD, motherboard, RAM, ect? Would I be better off just buying something assembled and used? If so, what would you all recommend?

Thanks in advance.

The software setup is way more important than the hardware setup.

In principle, while developing your probably don’t want to develop on the server and you’ll want to be able to run the majority of your project code on your laptop/desktop, and then build/deploy to a container of some kind either docker or k8s.

In general, proxmox + docker , or plain old combination of docker and KVM/libvirt is probably more than enough for things you described. (e.g. your typical $/core AM4/threadripper/xeon with 64/128G of ram with maybe ecc as a bonus).

But, if you plan to home lab microservices, you’ll probably want to give full fat kubernetes a spin at some point, and this means at least 10 VMs for a single cluster with non trivial ram and storage on each so you can run nodes and move your workload in there … and probably more than one machine.
For small projects the amount of overhead k8s has is scary for starters the assumption is you’re running at least 3 copies of everything and to access storage you don’t want to use local fs or nfs or fs in a thin provisioned VM image, you’ll want rook/ceph which means a daemon per node + some other management stuff.

It’s perfectly possible to do this in VMs it’s just not really recommended. However, for learning, you have minikube.

Personally, I need something like this too, I was hoping for 5000 series TR to drop, but might build an AM4 5900x / 128G ram / 4x18T proxmox host instead.

Do you have a specific need? other than “full stack/run everything”?

What’s your budget?

1 Like