With the average price of a new server today being 20k or higher most of my SMB clients can’t afford that. I have been experimenting with several Minisforum UM870 pc’s running Xcp-ng connected to either TrueNAS or a Synology NAS via 10GB network. This thing screams and is a small fraction of the price. I know it’s crazy to use a $600 mini pc as a business server but, I am thinking, I can stack several of these and if one dies, toss it and replace it. For what, a few thousand dollars instead of over 20K, anyone see a reason not to use this type of setup at a client?
I understand what you are saying, but personally, I would be hesitant to rely on that type of hardware for business-critical infrastructure. There are also a number of compatibility concerns with Minisforum systems when used as hypervisors, particularly around CPU and networking support.
For example, XCP-ng does not currently support Intel processors that use a hybrid P-core/E-core architecture. Some AMD-based Minisforum systems have also encountered networking driver compatibility issues. The MS-A1, for instance (and I realize you were specifically referring to the UM870) does not provide native support for Intel NICs.
In my opinion, you are likely to encounter more challenges than the initial cost savings justify. At least in my case, the potential troubleshooting and ongoing maintenance headaches would outweigh the benefit of using lower-cost hardware.
From a cost perspective, I also would not necessarily recommend purchasing brand-new enterprise hardware. There is often excellent value in the used market, particularly through eBay, refurbished enterprise equipment vendors, or even by reaching out directly to businesses that are refreshing their infrastructure. That approach can provide much more suitable and proven hardware without requiring the cost of purchasing new equipment.
This is just my opinion.
1 Like
I really appreciate your opinion! I am not married to the UM870, just the basic idea of a cluster of much less expensive pc’s where you can afford to toss one if it goes bad.
Are generally what I recommend, specifically Dell servers as they are easy to buy and get spare parts for that you can keep on a shelf and you can get them with lights out management. But I do imagine there is a near future where boxes by companies like Minisforum may become an option.
I simply would not run these units in regular performant mode unless I had removed most of the covers and had external fans blowing over them. Otherwise you might find your losses build up very quickly
That’s why I never buy them as I don’t expect them to last, used enterprise is the far superior option, only downside being power draw and noise
You can buy a 2ru server with single cpu (12cores), 32GB of ram, and about 18TB worth of storage (after configuration) for around $6500 each, and that’s with dual power supplies. Go up in ram and the board I bought jumps to 128GB and requires 4 modules. It is single or dual socket CPU, so you could run it up much higher if you needed. These are the new storage servers I have coming in soon for my VM storage, and probably a few other storage needs since it is “so big”, I need about 2TB for my VMs, 4TB is nice overhead, but the only drives I could afford were 4TB and I needed 6 of them to make a decently recoverable array (and one drive on the shelf or as a hot swap spare, not sure yet).
If you are buying used, the only requirement is that it has UEFI boot. This is becoming a requirement for many operating system. Secure boot is also a thing you may or may not want to have, as well as a TPM 2.x module. With the number of enterprise Linux that now support secure boot (without a shim), it’s getting easier to recommend using this and having the TPM 2.x module on board.
Another choice I might not really recommend, HP T740 or T755 thin clients. I’ve run mine pretty hard in my lab, 24/7 and they seem to be fine. I would need to know your use case before I would lean on these as a possible choice. Low performance, lots of heat, up to at least 90 watts draw at full go (AI Inferencing on one of mine). A basic AD or some other low-ish power need? Probably. XCP-ng for a few VM’s and make a pool out of them with either 10g or 25g NIC installed, yup, that works really well depending on load. This is my lab, I have 9 of them and a DIY NAS with Truenas to act as shared storage.
Does that run lab run VMware, yes it ran vSphere 8 really well, but never got around to trying vSan.
Does that lab run Harvester HCI, yes, but you can notice the overhead from this system. Not good for production, but good enough for a lab. Nutanix would probably be similar, and Openshift would definitely be similar. Proxmox apparently works really well on a small cluster of T740 ot T755.
Most of my clients are SMB with a single Dell PowerEdge server under ESXi running an AD server and a SQL server. Some have a third Windows server for Veeam backup. That is the cookie cutter all my clients get. Horsepower varies based on workload of course. Lots of discussion on new hypervisors but good or bad I have chosen Xcp-ng for all new servers. So far it runs on anything I have thrown at it with no issues.
Where do you all recommend I purchase the used enterprise hardware? It seems kind of strange, I mean, I am replacing a Dell PowerEdge (or whatever) server at a client because it is getting old so I go out and buy someone else’s old server they got rid of because of the same thing? That seems to be a very common recommendation, but it seems strange to me.
Age alone isn’t necessarily the deciding factor. If the hardware can comfortably support the client’s current workload, has enough headroom for future growth, and is still reasonably serviceable, then used enterprise hardware can be a very good investment.
There is also a meaningful difference between older enterprise hardware and newer lower-end hardware. Enterprise servers are designed for reliability, serviceability, redundancy, and sustained workloads. They are built to run continuously for years, and in many cases a well-maintained used server can still provide a better overall platform than newer hardware that was never designed for that type of workload.
For me, the key is not simply how old the server is, but whether it still makes sense from a performance, support, power consumption, parts availability, and lifecycle perspective. If I can reasonably expect it to handle the workload for another five years, I would have no issue deploying used enterprise hardware.
1 Like
Just had a dell 2950 die on me at a client happily chugging away as a pmox node until a few days ago… was released in 2006, now that’s something 
I’ve just purchased wherever, eBay etc… my criteria has always been as long as it has loads of ram already installed and cheap used cpu upgrade paths. But I always a spare leaving the second one in a cupboard somewhere at the clients
Very few SMB’s require anything more
I actually like the idea of using multiple low-cost nodes rather than putting everything on one expensive server. The part I’d be most concerned about is long-term reliability and how much extra time goes into troubleshooting hardware/driver issues.
If you had to choose, would you prioritize a few cheap nodes for redundancy or one proven enterprise server with better support?
That’s why I always buy an identical spare or two, they are usually sold in x quantities when sold on the used market
I have a customer running on several old dell purchased several years ago !!! There are 2/3 in the server room lying their for spares etc… there were more 
hi, you have not said what applications will be loaded onto the servers but I like your idea, with the exception of email and database/rdbms. email I would simply use Proton, and databases I would recommend using Dell, the rest of the systems, go for it, buy a couple of spare units and use UPSs.
What about just using the Cloud? The Grid?
Last server I built was in Feb 2025. I started with this ASRock barebones kit: https://www.neweggbusiness.com/p/9b-16-775-111?Item=9b-16-775-111&_gl=1*1tcejky*_ga*MTU3NDQxNDkwLjE3ODgyNzg4ODI.*_ga_RM171ETZBN*czE3ODgyNzg4ODEkbzEkZzEkdDE3ODgyNzg5MDEkajQwJGwwJGg0OTI0ODM1ODM.*_fplc*JTJGNTcwZDQlMkZpN0RQJTJCa2JsWGhFYVJJV1NkMUhSNnV3OWxTV1JuN1ZRcXFsNmlWQzJuRm05RDVQV0h0d0t3eWtwODBROExaVTdYJTJCOW54UE00NTRkdHRTNFF6Wm14TXJYaGcxellwQSUyQnhNbiUyQmljeXlYa1l1WjlZa2wweGlFN1dRJTNEJTNE
I purchased NewEgg’s 3 year warranty, so that made it around $1565 at the time.
I outfitted with 64GB RAM, AMD EPYC 4446P Processor, 2 x Samsung PM893 SSD (RAID1) for OS, and 2 x Samsung PM9A1 2TB PM9A1 NVMe SSD for data. Running MS Hyper-V with 2 VMs…1 DC VM, 1 File / Print VM. Not Enterprise, but I consider it mid-level and perfect for SMB. It performs very well. Server build at that time was around $3200 + MS Licensing, around $1526, Total: $4724. I know prices and availability have changed moderately since then, but even at double the price you are at $9400, and in your build you wouldn’t have the MS Licensing.
Lonnie