As I plan my first dedicated homelab build, I’m seeking to address several limitations in my current repurposed Lenovo Y710 Cube setup. This system features an i5-6400 processor, 32 GB of DDR4 RAM, a DRAM-less 1 TB Crucial BX500 boot drive, two 12 TB Exos drives passed to TrueNAS, and a GTX 1060 6 GB GPU. It currently runs Ubuntu with approximately 20 Docker containers, TrueNAS Core, and several LXC containers including Pi-hole, Unbound, DNSDist, Jellyfin, and FileFlows. However, I’m facing constraints including the motherboard’s 32 GB RAM limit, lack of IOMMU support, consistently high memory usage around 80-90%, IO lag from the DRAM-less SSD, and limited upgrade options.
My objectives for the new build include improved NAS functionality potentially using Unraid, the capability to run a Linux gaming VM with Sunshine/Moonlight for AAA single-player titles on a 4K television, and expandability in terms of RAM, PCIe lanes, and storage. Additional considerations include potential integration of a single-board computer for firewall purposes, quiet and energy-efficient operation suitable for a Northern European apartment, future machine learning capabilities, enhanced networking with VLAN support, and a flexible platform for homelab experimentation.
For components, I’m evaluating a Ryzen 7 9700x processor paired with an Asrock B850 Pro RS motherboard, 128 GB of DDR5 memory, and an Nvidia 5070 Ti GPU. I’d appreciate guidance on whether the Ryzen 7 9700x provides adequate future-proofing or if I should consider Ryzen 9, Intel i7-14700, or higher-core alternatives. Regarding media transcoding, I’m curious if AMD’s integrated or discrete graphics would suffice for Plex/Jellyfin in Proxmox, or if NVIDIA/Intel Quick Sync remains the more reliable option. I’m also wondering if 128 GB of DDR5 RAM might be excessive, where I could potentially reduce costs in this configuration, and whether to prioritize motherboards with integrated remote management or consider separate KVM solutions.
I’ve been running a similar setup with a DRAM-less SSD, and that IO lag you mentioned is real—it became a real bottleneck for my Docker containers until I switched to a drive with a cache. Your plan to move to Unraid for better NAS functionality and a dedicated gaming VM sounds like a solid path forward; I did something similar to host a Windows VM for Steam streaming. Are you leaning towards a specific platform for the new motherboard to get those extra PCIe lanes?
Thanks for sharing your experience—switching from a DRAM-less SSD really does make a noticeable difference, doesn’t it? For the new build, I’m leaning toward an AMD AM5 platform, as it offers solid PCIe 5.0 lanes and solid IOMMU groupings for GPU passthrough, but a used Epyc or Threadripper setup could be a cost-effective alternative if you need even more lanes. If you’re exploring similar options, checking the hardware compatibility list for Unraid or Proxmox is a great next step to ensure smooth VM performance. I’d love to hear how your Steam streaming setup is working out!
I remember hitting that same 32GB RAM ceiling on an older system, and it’s frustrating when your container stack starts to feel the pinch. Your plan to move to a platform with better IOMMU support for a gaming VM is smart—I did something similar with a 3060 Ti for Moonlight streaming, and the difference in responsiveness was huge. Are you leaning towards a specific motherboard chipset for the new build to get those extra PCIe lanes?
Thanks for sharing your experience with the 3060 Ti and Moonlight—that’s exactly the kind of responsiveness I’m hoping for! I’m currently leaning towards an AMD B550 or X570 chipset for the new build, as they offer solid IOMMU groupings and those extra PCIe lanes for the GPU passthrough and additional storage controllers. If you’re curious about specific board groupings, the Arch Linux VFIO guide is a fantastic resource for checking IOMMU compatibility before you buy. I’d love to hear if you have a particular motherboard model you’ve had success with in your setup.
I’ve been running a similar setup with a DRAM-less boot drive, and that IO lag you mentioned is so real—it really sneaks up on you once your container count grows. My own next step is migrating to a used enterprise SSD for the host OS, which made a huge difference in a friend’s Proxmox box. Are you leaning towards a specific motherboard chipset to get the IOMMU support and more PCIe lanes for your gaming VM?
Thanks for sharing that—it’s reassuring to hear I’m not alone with the DRAM-less SSD lag, especially as the containers pile up. I’m leaning strongly toward an AMD platform, like a used Ryzen 7 or Threadripper paired with an X570 or TRX40 board, for the solid IOMMU groups and plentiful PCIe lanes to dedicate the GTX 1060 cleanly to a gaming VM. If you’re curious, the ServeTheHome forums have some great build logs for exactly this use case—I’d love to compare notes once you pick your motherboard.
That seems like overkill. I use three used Intel NUCs for my Proxmox cluster, which are both affordable and reliable.
For your first homelab build, I’d recommend Intel over AMD based on my experience. The main advantage is transcoding performance. If you plan to use Jellyfin—which I suggest trying before Plex—Intel is significantly better. As someone who prefers AMD, I still believe Intel is the superior choice for both CPU and GPU in this case. If Intel isn’t an option, consider Nvidia instead, as AMD has received less support for transcoding and has fewer advocates.
I’m not sure which clients you’ll use for media, but our Apple TV, iPhones, MacBooks, and PCs don’t require transcoding with Plex. I have it turned off entirely because my 7th gen i7 was getting too loud, and it works fine without it.
I don’t need live transcoding at home, but I run Fileflows workflows that use VMAF compression and transcoding before storage. Due to CPU limitations, I can only handle one transcode at a time.
I also plan to expose the service, which complicates things since I won’t have control over client media access.
For media transcoding, consider an Intel processor.
Consider an A380 or A310 instead of an Intel iGPU for better performance.
Given the higher energy consumption and potential issues with the i7 14700, you could easily add an Arc GPU or reuse your old GTX 1060 if the AMD iGPU isn’t sufficient. My main concern is ensuring the CPU doesn’t bottleneck your intended use, as it does with your current setup.
Most modern Intel CPUs include dedicated units for encoding and decoding. Even the N100 and N305 models handle this with virtually no CPU utilization.