AMD Threadripper CPUs are surprisingly efficient for home labs

When one mentions an AMD Threadripper CPU and efficiency in the same sentence, it can draw some raised eyebrows. How on earth can a 64-core beast like the Threadripper 9980X be anywhere near efficient with a TDP alone of 350W? You can expect this CPU to draw up to 400 watts of power through the socket at stock settings, but after using it in place of a few used enterprise rack servers, I've discovered it to be incredibly efficient, even compared to a few mini PCs.

Most would automatically assume an AMD Threadripper, being a high-end workstation processor, would suck so much power that it would render it almost useless for home lab deployment, should you have energy usage in mind. Interestingly, I found it to handle various workloads well for self-hosting content and apps. It's actually surprisingly efficient when considering just how much performance is available on the platform.

Threadripper is great for home labs

More cores, more applications

Self-hosting apps such as Jellyfin and Immich won't require a behemoth such as the Threadripper range of CPUs, but once you start throwing in virtual machines into the same mix, things can get a little overbearing for even more capable consumer (and used) enterprise hardware. That mini PC with an Intel or AMD mobile processor will be perfectly fine for running a few containers, but running local language models (LLMs) will be out of the question without a dedicated GPU.

I had an array of enterprise servers running Proxmox, each assigned a few containers and VMs to run when required. This was quite the setup and served our demands well; it also fed into my small obsession with sourcing used enterprise gear on classified websites. The primary drawback to this approach, aside from proprietary parts and sometimes difficult troubleshooting, was the power usage. These servers can crank up to a good few hundred watts per system.

You'll often find advice from more experienced system admins and home lab enthusiasts to consolidate systems into smaller, more compact, and generally more efficient platforms. Mini PCs and consumer-grade components are a great way to achieve this, but so too is building a single PC with a more powerful foundation. Enter AMD Threadripper with its workstation levels of performance at more reasonable prices (and power requirements) than EPYC.

Home labs are built with a balance in mind for cost, performance, and versatility. I use mine for running self-hosted services to save money each month, but it's also a great playground for testing new things, learning, and handling pretty much all our smart home and security needs. The most appealing specification for AMD Threadripper CPUs, specifically the 9970X and 9980X from the most recent generation, is the high core and thread counts.

I can run LLMs, Frigate, Jellyfin, Nextcloud, Immich, and countless other VMs and containers without causing strain to the CPU, which is fantastic for running everything on a single platform to save on cost. And because I'm not running a datacenter where every second counts, should the system encounter issues, I can source new parts or quickly put together a replacement using other parts and deploy a Proxmox backup to restore everything with only a brief outage.

Threadripper is energy efficient

It uses less power than you think

AMD has worked on efficiency for the past few generations of Threadripper. The 9000 series is particularly frugal when it comes to sucking power from your motherboard, even at near-idle loads. The Lenovo ThinkServer SR250 V2 is a great rack system, but even that idles at around 70W or so. The AMD Threadripper 9980X with an Asus board, SSD, two Nvidia GPUs, and 128GB of RAM sits at around 130W (measured at the outlet). That's around half of what my primary PC uses (including monitors).

Firing up Frigate puts the RTX 3060 Ti into full swing. This makes the power draw climb to around 250W with five IP cameras connected and detection conservatively configured for best results without taxing the system too hard. For a full security system combined with hosting all our apps and services, we're looking at around $2 per day in electricity costs for the AMD Threadripper Proxmox system, totalling around $720 per year.

We don't pay for Spotify, Office 365, and many other subscriptions, so we're still saving money each year, yet I enjoy the benefits of self-hosting and having a powerful enough system to have a play around with various software configurations. Should I finally get around to installing a solar array on the roof and working in battery storage, we'll be looking at considerably less draw from the grid, making Threadripper even more efficient in that regard.

The best part about using a system with enough headroom for future growth is that we can calculate how much power we'll be using when deploying more services and apps. Unlike using mini PCs and rack systems, where we'd need to factor in the power requirements of an entirely new system, the Threadripper-powered server will need just a few watts more to run more containers or VMs. If I want to run some network sniffers and try out some other things, I won't have to worry about resource allocation.

It's not all great news for home labs

You've read through my reasoning for using Threadripper in a home lab, but there are plenty of reasons why this may not be the best idea. The first (and foremost) common concern is initial cost. AMD Threadripper processors are not cheap, costing multiple thousands depending on the SKU. It's possible to pick up a used Threadripper CPU, but I'd recommend the most recent 9000 series if possible. The 9960X will set you back an MSRP of $1,499 for 24 cores.

Then there are the requirements for powering and cooling the CPU alone. You'll need a decent PSU with the best efficiency rating and preferably an AIO liquid cooling kit, though the more expensive air coolers will work fine with lower sustained loads. These can be balanced out by ECC memory support, plenty of PCI lanes, and impressive versatility for networking, virtualization, and media processing. If you can think it, these CPUs can run it and it makes for quite the compelling home lab.