I'm currently obsessed with building out a home lab with a server cabinet, some rack systems, and plenty of equipment that's either overkill for what I task it to do or is a little too much for the home. It's possible to go with a small collection of mini PCs or scour classified websites for great deals on used enterprise hardware. I tend to do just the latter, which ends up requiring a fair amount of space.

The only drawback to using such hardware is although they are compact and relatively powerful, these are often systems with generation-old hardware, sometimes a few series back. Many modern apps and services often fully saturate available system resources, meaning you can end up with a few servers, which is precisely what I have. That's when I decided to do something about it.

I've begun the process of upgrading some of the CPUs within these servers. This is actually a good cost-effective solution to the problem, sometimes doubling the number of cores and threads I previously had per machine. But there's also another route I could take by replacing most of the entire server cabinet with a single system, rocking an AMD Ryzen Threadripper 9980X.

AMD's all-new 64-core monster

How this impressive CPU came to be

We've already reviewed the AMD Ryzen Threadripper 9980X and gave it a high rating, thanks to its impressive number of PCIe lanes and quite frankly ridiculous memory bandwidth. Zen 5 (codenamed Shimada Peak) is the architecture powering Ryzen 5 to Ryzen 9 processors, as well as all the 9000 series Threadripper chips. What you get with these Threadripper chips are the best bins, meaning they're top-notch and designed for the heaviest of loads, represented by their sheer size.

TSMC's efficient 4 nm process delivered notable IPC improvements over previous-gen Threadrippers. What we have with the 9980X is essentially eight Core Complex Dies (CCDs) stuck together with 256 MB of L3 cache, creating a massive processor. Each core has its own L2 cache for storing as much data locally as possible, though a whopping 80 PCIe lanes and quad-channel DDR5-6400 support will see data transfers happen across the sTR5 platform in a flash.

In terms of specifications, we've got a base clock speed of 3.2 GHz and a maximum boost of up to 5.4 GHz, which may not sound like much, but this thing has 64 cores and 128 threads. That's quite the number, quadrupling that of the 9950X, AMD's flagship consumer-grade CPU. With up to 350W of power to work with, there's almost nothing this Threadripper cannot do and it's not even the flagship SKU!

Consolidating a busy home lab

Less is more ... sometimes

I've made good use of the trusty Lenovo ThinkServer SR250 V2. It's a cracking machine with a powerful Intel Xeon E-2336 processor and 6 cores. On this server alone I have Immich, Jellyfin, and Home Assistant all running happily, along with a few other minor containers and VMs. Everything is no match for the CPU, clocking in at around 15% at idle or low demand. Even when transcoding something through Jellyfin, I'm only seeing up to 30% load.

But then I have another machine dedicated to running more VMs and containers since I want to leave ample headroom for each service. That's where consolidation (or the need to) comes into play. Throw in a third server, a Dell PowerEdge R210 II, for running Frigate, and a fourth doing something else, and you can see how things can quickly escalate. Instead of having five (or more) servers humming away, drawing power, why don't I cut it down to just two?

This is what the Threadripper family of processors can do. They can either be used to brute force your way through heavy loads or consolidate numerous tasks onto a single platform, and the latter is precisely what I had in mind. Using a server cabinet full of racks is great, but it's a hassle to manage, even with a KVM. Then you have multiple Proxmox nodes or individual installations. It's loud, hot, and a pain to work around.

Having just one or two machines running cuts costs down and while it can produce a single point of failure, bringing down everything running on the node, Proxmox has a robust backup feature and ZFS is excellent for snapshots. Plus, it provides me with the means to use one of the existing servers as a backup machine to hold a copy of everything. Following the 3-2-1 back up rule, of course.

Proxmox brings everything together

The undisputed king of virtualization

Proxmox quickly became my de facto virtualization platform largely thanks to how good a hypervisor it is. Taking full advantage of the lower cumulative power draw, thanks to reducing the number of running servers, the Threadripper 9980X is a perfect match for Proxmox, running LXCs and VMs. Once fully optimized, even though the 350W for the CPU alone should sound some alarm bells, the CPU is that capable that it barely draws anywhere near as much juice.

The newer architecture is miles ahead of older Xeons I relied upon in terms of efficiency. For each task assigned to one of the 64 cores, the Threadripper completes it quicker and using less power, resulting in savings across the board. There's also less network load, which is often overlooked when configuring a home lab with a server cabinet. Sure, 10Gb links between switches is more than enough to provide ample bandwidth, but a single Threadripper system removes that need altogether.

Thanks to what Proxmox can do, I can use this Threadripper build for KVM-based virtualization, LXC containerization (hello, Docker!), ZFS storage, clustering, and more. It's incredibly versatile and can easily run media servers, developer environments, game server instances, networking support, and more with snapshots and remote backup all included. It removes the need to cluster less-capable nodes, and is fully proofed for the future with plenty of PCIe lanes.

As covered in our in-depth review of the 9980X, the performance is quite frankly ridiculous. I was contemplating maybe going down the route of building a compact server with a Ryzen 9 9950X, but the Threadripper 9980X absolutely blows that CPU out of the water. It should, costing considerably more, but it goes to show just how capable these chips are and my workload is just right for such an application.

The highlight for me is killing off my dedicated LLM system, which is only tasked with interfacing with an RTX 4060 Ti for running Ollama and Open WebUI. I can now bring LLM deployment into the fold without worrying too much about CPU, RAM, and GPU resource management.

The home lab must grow

It's always a work in progress

The home lab is never finished. It's a hobby, not a project. There's always something else to tinker with or replace and I fully expect I shall return here at some point once I find other ways to use the older servers or even pick up a good deal on another Threadripper. There's something about running all these powerful hardware right at home with so much potential at your fingertips.

It's often more about what your home lab could run, not what you do run. Have you reached the point where you began consolidating hardware within the home lab? I'm interested in learning about your stories for building out your home-based server infrastructure.