Threadripper vs Epyc for Unraid home servers Compare Threadripper and Epyc options tailored for Unraid setups.
Threadripper vs Epyc for Unraid home servers Compare Threadripper and Epyc options tailored for Unraid setups.
Hello everyone, I just found out that a Ryzen 1700 isn't quite right for my setup when it comes to PCI-Express connections for my unraid server. Since I need at least two video cards—both for VGA output and decoding—I’m running low on PCI-E lanes for a third card in a virtual machine. The simpler choice right now is switching to an APU, but these are hard to find. I’ve been considering Threadripper and Epyc models, and it seems that Epyc systems can be more affordable while still giving more PCI-E slots than Threadripper. In return, Epyc might sacrifice clock speed compared to Threadripper. There are other considerations too, like the size of the mainboard, cooling options, and form factor. I’m also curious about how this affects hosting game servers—I remember seeing a big jump in performance with the 4300 (with piledriver) setting it to 4300MHz and boosting memory speeds for games like Factorio and Ark. Would that difference be similar between Threadripper and Epyc? Also, could you share your thoughts on whether clock speed really matters for game servers or VMs, and any other practical points such as cooling or form factor? Lastly, I’ve noticed others using similar builds for home use lately, and I’m getting more into heavy hardware like advanced networking and servers. Thanks!
Noted, the performance depends on how well clients handle cores and clock speed often wins over raw count. If your tasks fit GPU acceleration, prioritize that. You don’t need a massive setup—around 1660 or 1650 should handle video rendering well, and many engineering tools support multiple render nodes to boost a single project.
It's working well with the 1050Ti for shinobi and plex, but I've seen my requirements increase rapidly because I'm eager to explore more games. That's understandable. More games are indeed adding more cores to their main threads, though it hasn't quite reached the level I'd prefer. Let's check what others say.
SP3 employs identical mounting methods as TR4, which shouldn't pose a significant concern. For heavy use of high-capacity registered dimmers, consider directing airflow toward the memory module. A fan aimed at the VRM is also advisable. When selecting devices like the EPYC 7443P (or 16-core variants), performance may be more limited, but each unit will likely underperform compared to others. To make full use of the PCIe bandwidth available on the EPYC platform, it’s wise to anticipate using extensions and U.2 to PCIe adapters. Keep in mind the availability at purchase time. Intel Ice Lake Xeon models are recommended; their entry-level options offer decent value despite only providing 64 lanes of PCIe.
Thanks for the input Goldenlag. The feedback was valuable. I'm already considering some risers and planning a 4U or home build setup, which gives me room and spacing. A 64 PCIe lane should cover it. Right now I expect to stay under x8 on a graphics card—my GT710s are X1 models. That means the 1050 Ti on an x8, possibly my current 1080 on an x8 or x16 if 30 series arrive at good prices. I have one SAS controller with four lanes, bringing my total expansion slots to between 22 and 38.
Feel free to use a budget used 40Gbit card and connect it to the server’s PCIe slot for external storage. You can also look for affordable U.2/PCIe 2-bit MLC/DLC SSDs. These options offer strong endurance and consistent write speeds. If you’re using a standard case with good airflow, consider adding a few 120mm to 40mm fans (up to 92mm works well). Server boards often lack effective fan control.
Who understands what the future holds? …Options, options, not quite excited enough yet to launch a homelab where I’d actually need this. But yeah, options. The current setup relies on two M.2 NVMe drives for cache (which also explains why I’m running into my problem first), giving me most of the speed required right now. But I hear you. For long-term storage, I’m using four 4TB WD Red drives in an array. When Unraid supports a second array, I plan to upgrade with two Purple drives for CCTV recordings, which are currently cached before being written to the array. For the VMs, I’m mostly using leftover SSDs I don’t have another purpose for at the moment—plenty of fans around, kind of a hoarder in that sense. A nice note about the fan controller? Maybe I still have an old 5.25 fan controller somewhere, so I can at least manage the fans. The server’s location isn’t perfect when it comes to temperature.
You can purchase an inexpensive 5-10 dollar PCIe video card from eBay that only outputs VGA, and trim the connector until it fits a PCIe X1 slot. This design allows most PCIe devices to function with just one lane, so inserting your card into a PCIe X1 slot should work fine. Once you have the first PCIe X16 slot ready, you’ll likely find the bottom X16 port available for devices that truly need it. The bottom X16 slot also offers four PCIe lanes, which is sufficient for handling multiple encodings and decodings at once. A card meant for hardware decoding won’t gain much advantage from sixteen lanes or direct CPU connection—four lanes are more than enough for simultaneous processing. It’s unclear why a single PCIe lane wouldn’t suffice for both VGA output and encoding, but it seems like you’re overlooking the performance benefits of a dedicated solution. I noticed some comments suggesting a M.2 adapter in the X16 slot could work if your motherboard supports bifurcation, allowing four NVMe slots. You might also use basic PCIe X1 to X16 riser cables to connect the encoder/decoder card without cutting the connector. Even a single lane delivering over 900 MB/s should easily support hardware encoders on the card.
Gday Marius, I know you can trim parts of the slots to fit them, but I’m not keen on that. The SAS controller requires four lanes, and it functions at x8 since the 1050ti shares the same bus. The other x1 slots are on the chipset, which only has four lanes available—mostly occupied by two NVME drives. At the same time, passthrough for Docker isn’t working yet. When Unraid boots, it demands a GPU, and I haven’t managed to get full control to make it headless. Right now, I’m stuck between assigning it to Docker or a VM, especially since it’s not being used as a VGA output for Unraid itself. This post is just about exploring options, as I’m hitting a roadblock in fixing the issue. There’s also another problem: my internal setup breaks when I add a second GT710 (see arrow). My SAS controller seems to malfunction. Could you explain why? I’d like to try and resolve it. From what I read elsewhere, the four 3.0 lanes going to the chipset are downgraded to eight 2.0 lanes—giving four for NVME and three for the x1 slots.
Consider using an older dual Xeon E5 first-gen processor—it's affordable when refurbished. For instance, the Xeon E5-2690 with 8 cores offers a good balance of performance and cost, often available around $50. Pairing two of them gives you access to 80 PCIe 3.0 lanes. Used DDR3 ECC memory sticks are also very budget-friendly.