I had something planned several years ago, but ECC is a personal “need”, so even older hardware is out of reach. I wanted to build a server for everything, play with VFIO on Linux and still have a decent system for gaming in a 32U rack. As much as I don’t like to political, it’s most definitely because of Trump. 
So what is your dream homelab build if money were no object?
Can’t say it’s just ram any longer. SSD, NVMe, HDD’s are all affected as well. Big tech is trying to push self-hosting away from people and companies. I don’t think anything current was really affected. For me it’s all stuff I plan on doing later like expand NAS storage or NAS RAM, VM server RAM, more ssd/nvme storage. Think the main hit I took was purchasing a replacement laptop.
It’s literally steel prices that stopped me from going vertical; rack and transplanting my current computer to a rackmount case. Hundreds of dollars that I didn’t have to begin with added on top.
I had an EPYC 8004 build planned. At the time, I could always have built a separate gaming PC.
My ‘HomeLab’ was going to be the coolest off-grid fully automated setup that was self reliant and energy independent. Now, I can’t afford to finish building my home because of the collapse of this damn country. So now I’ve got to downsize and sell it all and move to where there are like minded people and good community (the sticks where I’m at now are full of those ‘useful idiots’) so my kid can at least know reality and ‘decent people’.
I’ll build again, just not in this country. – For the time being however, it’s fun making old hardware do new things they were never meant or designed to do. All us ‘poors’ can do is look out for each other now as I fear it only gets worse from here. .
It is now more what can I do with what I have with some adaptors, and existing horded parts.
I did get an intel B60 to play with some SR-IOV in proxmox. If it works, I’ll have 2x 3060 12gb to rotate out.
I still have months worth of tinkering to do before I catch up with the list I have, then adapted, buried, retrieved, filed in triplicet, lost, found, and then finally acted upon.
Waited too long to buy a big pile of 12TB ironwolf drives, they are $430+ these days and unavailable, I at least got 12 of them in a NAS for $230 each not that long ago and they were always on the shelf back then (along with the $300 4TB NVME drives that were on the shelf, or before that the 8TB SATA Samsung QVO SSDs for $300-400). I have a few 20+ TB drives but went with 12TB for the “big” unfinished NAS as a happy middle ground for potential resilver time and decent performance, plus the cost.
I missed out on buying some things, but got some other stuff at the old prices so I’m not too upset on what I didn’t get, some projects will go unfinished, others will cost a lot more. We will probably never see prices like we did due to inflation plus the insane demand that won’t vanish even if the public facing AI bubble pops tomorrow.
This AI hype and circular economy ramp up was going to happen with or without trump, it was a massive machine with a LOT of momentum even years back, things wouldn’t be different by any appreciable amount under someone else even if some details would be different.
That’s genuinely lucky.
@MadGoat Do what is best for your kids (MAAAAAA! I hope this joke finds you well 
), first, of course. Without sounding like a broken record, try to keep them stimulated with STEM, as much as you and the school system in your area are able to allow. I am sorry that that happened to you. 
I forgot that I can’t edit. Yay.
I think that if any of us knew the future, we would all shit ourselves, but I would bought storage a lot sooner, too.
Yeah I got affected by this as well (it kicked in right as I was starting to execute my plan)
My main NAS I finished building I think April this year has one 2Rx4 32GB stick of DDR5 ECC and that one stick cost £480 at the time (it is now >£1000!).
My original plan for this server was for it to have 256GB. But that’s out the window.
I also wanted to go all-NVMe and ditch my aging and slow 870 QVO SATA SSDs. But that’s out the window.
I was also going to build 2 more NASes for different storage and backup roles, one with 256GB of DDR4 ECC and one with 64-96GB of DDR5 ECC. There was going to be about 100TB of capacity between these two other boxes. But those are out the window.
Finally, I was going to build a GPU compute box for messing around with local AI and for remote video rendering. Multi-GPU, blazing fast CPU, blazing fast RAM. But that’s out the window.
I had just saved up enough to buy a rack and a server case.
That is all very nice, though. Nice plan.
For the record, I had a kitten grace me with her presence instead. Still drained my entire savings when she got sick, but totally worth it. 
I can’t edit to say that I get it. I wasn’t trying to be an arse in my response. I apologize if it came off that way.
Happened to do a build out right before it all hit, so reasonably well positioned.
NAS with ~4TB redundant flash & ECC. Plus a janky 2nd NAS with 10TB redundant rust.
Could always be bigger / better / fancier but it’s decent enough.
Desktop on the other hand has me rather nervous…that’s due for upgrade in next year or so. 
My home lab may have had 128G in it on my AM4 box after I upgraded to AM5.
But not buying a heap of new DDR5 in this economy!
Well, I kind of have my “dream homelab”, but it is getting a bit long in the tooth now in terms of specs. It still runs though.
My setup has always been geared towards self hosting game servers and development related stuff and consists of three compute machines, two storage machines, a network management server, a router, a couple of switches and a couple of UPSes.
Previously this was all DDR4 based:
the compute nodes (Dragon, Rhino and Mustang) all had SATA SSDs and most of the VMs ran via iSCSI and 10GbE and all had >8 core AM4 CPUs (5800, 5800X3D and 5900), as well as 128GB ECC UDIMM RAM in their final iteration. (over the years the RAM kept getting upgraded)
The primary NAS (Elephant) was a dual 2033-3 setup with 128 GB RAM, a quad 10GbE card and went from eight 4TB HDDs to eight 4TB-ish SATA SSDs.
The secondary NAS (Moose) is purely for backup of the primary NAS and still suffices. It is based on the Xeon D-1541 and has 32 GB of RAM. It has six 16TB spinners for storage. (it stared with 3, I kept adding one every couple of years) Since it does not need to be very performant, this will likely be fine until the motherboard dies of old age or something.
The network management machine (Suni) hosts[1] VMs that are outside my cluster and it was a Xeon D-2123IT machine with 128 GB of RAM and a couple of SATA SSDs.
The router (Dik Dik) is/was a Epyc 3101 machine with 32GB of RAM and dual 10GbE. I have bought a 2nd hand UniFi Enterprise Fortress Gateway as its replacement in the hope that it will be able to route 10GbE at line speed with filtering, but I dislike the requirement that you need a cloud account to use it, so I have not put it into use yet.
The switches are also Unifi: a Ubiquiti UniFi Switch Pro Aggregation and a Ubiquiti Unifi Switch Enterprise XG 24 for the higher speed traffic and a Ubiquiti Unifi Switch Enterprise 24 PoE for the PoE powered devices and management interfaces.
A pretty nice setup for what I need(ed)
Last year local LLMs entered my life, so that has been the main focus of my spending recently. And there the RAM crisis did rear its head. Tapir, my first inference server, was built from the hardware that lived in my desktop up until that point, but which I put into a server case, as I upgraded the desktop to a (2nd hand) 7800X3D and 64 GB, which was still quite affordable early in the year.
So present day is a lot more about compromises than before, but with some balancing I am more or less getting by.
The second LLM machine is Wisent, and it did suffer. Ideally, I would have gone TR5 for that one, but that was waaay too exepensive.
I was thinking of lowering my power draw as well, as the NAS was drawing like 150W constant load. Becasue of this Elephant was eventually “upgraded” to a 7900X and 192GB of DDR5, with an Asus X870 ProArt motherboard, just before the RAMpocalypse hit. It did manage to lower the power draw noticeably, but the lack of connectivity on the X870 board was a bit problematic. At the same time I really wanted to upgrade my game servers to DDR5, so now the motherboard and half the RAM lives in Dragon now, I upgraded Rhino with another X870 ProArt and the other half of the RAM and changed Elephant to a TR4 system based on a WRO80 Creator 2.0 with relatively cheap 2nd hand DDR4 I managed to get my hands on. However, I also got an upgrade for the backplane boards, changing the interfaces from SAS/SATA to U.3, as well as four retimer cards, making it possible to go full NVMe in the future. I think this is probably going to prove very necessary, as I think that the Storpocalypse has put SATA out to pasture without much noise being made about it. (Try finding 4TB- or larger SAS or SATA SSDs, compare that to the availability of NVMe SSDs and you’ll probably get why I’m thinking this way) For a NAS DDR4 is likely good enough for what I want to get out of it in the foreseeable future. An upgrade to 8TB SSDs is on the: “some day, but not in the foreseeable future” list.
Since NVMe will (hopefully) be the future for Elephant, once storage prices become normal again, upgrading the core network to 100GbE is also on the radar, but it is still too expensive for me right now. Mustang also could use an upgrade, but it is not really necessary.
Ideally, all of the game servers would run with 128+ GB RAM and 4TB SSDs in mirrors, but that is also not going to happen for a while.
I would say that for me RAM is a bit of an issue but storage is a larger issue for me right now. If a disk dies I am in greater trouble than if a server has to go by with less RAM. Also, in the end the homelab is never “done”, it is just in the process of being upgraded to the next state 
¹ it died a couple of weeks ago and now I have to figure out how to replace what it did 
Footnotes ↩︎
All Good, didn’t read it that way.
I am lucky in the fact that I have a wife that tolerates what amounts to a mad scientist lab in the basement with parts, systems, and cat6 and fiber strewn about.
I was planned in building an AI Workstation, but when I realised the 256GB of DDR5 RDIMM alone took the entire budget I bought two DGX Sparks instead. In hindsight I’m really glad I didn’t go down the original route.
I am still debating the fate of my homelab. Everything is staying as is for now, but going forward I see things going a different way. Rather than having a single large server I will likely use a mini-pc to run any apps I need and then something low power to use as a NAS. My NAS is going to lose some capacity to gain a few extra drives. It is just prolonging the inevitable. HDD availability needs to improve in order for me to keep moving forward.