Quite often, I see people bragging about having a 2,000$ server in their closet… or trying to spec out a piece of hardware for running a plex server, etc.
This article is just a rant, about not considering the actual needs of the services you intend on running.
Many times I hear somebody say… well, go eBay a dell poweredge. Yes, it is true you can get a pretty powerful server for 200$.
A typical rack-mount server:
The above server contains TWO X5570 CPUs, each having 4 cores, 8 threads. You get remote IPMI (ability to manage the hardware remotely, if you OS crashes, etc…), 8GB of ram, etc. Its a great deal, right?
The catch is- Power consumption, and noise. If you don’t have an issue with this device drawing 350 watts around the clock, and sounding like a jet engine taking off during load, by all means, go grab it.
To note- 1U servers are notoriously loud due to limited space. 2U servers are generally much quieter.
A typical corporate-workstation
Now- for a more REASONABLE approach, you can instead…. get a dell or HP workstation.
Lets take this posting, and compare and contrast.
For one, your power consumption will be AT LEAST half of the poweredge server, due to only a single CPU, rather then two. You have the same amount of ram.
Overall- you will have less cores/threads, since you will only have a single 4c/8t processor, rather then two of them. However- you will have a FASTER single processor, which is 4 years newer, and clocked higher. If you have a massive multi-threaded workload, the dual xeons “might” be faster… but, in most cases, I would lean to the i7 being faster due to improved IPC.
Based on CPU Benchmark, the i7 will outperform both X5570s combined, while drawing 77W TDP, compared to 190W TDP.
Also- corporate PCs are QUIET. Companies will build cubicles with row after row of employees using corporate-workstations. A requirement- is the hardware needs to be quiet, for people to work!
PROs / CONs
What are the CONs to a rack-mount server?
- Power Consumption
- Space Efficiency (If you don’t have a rack)
What are the CONs to a workstation?
- No hot-swap drive bays
- No IMPI / Baseband management
- No “enterprise” features in general.
- Internal expansion room is usually pretty limited.
What are the PROs to a used rack-mount?
- TONS of CHEAP ram. You can get 128GB of ram and the server, for less then 200$.
- TONS of CHEAP CPU. An older xeon might not be lightning fast, but, you typically get two, or more of them- and they typically have 4 cores and hyperthreading.
What are the PROs to a workstation?
- Higher clock speed is sometimes very important for some workloads. (game servers, especially the single-threaded ones).
- Small, easily hidden. Quiet. (I have one hidden on my windowsill in my livingroom.)
- Typically- very easy to work on.
What are the CONs to a home-build?
- Well, it costed 500$ to make an extremely minimal server, (Talking about the 500$ server linked above). For 500$, you can get a pretty damn powerful rack server.
- You can pretty much always find a cheaper, used server on eBay.
- Unless you spend good $$$ on a motherboard, getting remote IMPI actually takes a bit of effort.
- Unless your home-build includes a supermicro server case & mobo, you aren’t gonna have 32 DIMM slots and 24x PCIe lanes……. unless you spend 2,000$.
What are the PROs to a home-build?
- You build it exactly the way you want it. You want small, and minimal? Fine. Want 20 3.5″ HDD bays? Fine!
- Newer technology.
- Is as fast or as slow as you want it.
How much CPU do you actually need?
One thing which gets mis-represented often, is how much processor you actually need.
More is better, right?
Sure, if cost and power usage is not a factor.
Take my personal server for example- It uses a 80$ quad core ryzen, without hyperthreading.
Most people would instantly look at this, and call it under powered.
Lets see what CPU benchmark thinks.
Well- Look at that. It draws less power then either of the above two options, and is faster at passmark’s benchmark. It is much newer, and offers improved IPC as well.
But- its only a quad core, so, it can’t run my applications!
Lets take a look at the docker services I am currently running on it.
- Mail relay to forward emails to mailgun.
- cloudflare ddns server
- technitium dns server.
- swag (essentially used to get let’s encrypt certs)
- unifi server
- esp home
- home assistant
- mosquitto mqtt server
- node red
- mysql server for home assistant (recorder, state changes, etc)
- next cloud (essentially, google drive/google photos combined into a wonderful interface… also- keeps backups of files from my various computers)
- mysql server for next cloud
- deepstack (does ML/AI for my NVR footage to only alert me when there is an actual PERSON walking around my yard)
- gitea – essentially gitlab. but, much easier to manage, imo.
- splunk – I log all of my network, server, and security data here.
I have a server 2019 server running Blue Iris. It captures from a handful of 5MP reolink cameras, 24/7.
So- with all of those services, I must be close to the limits of this quad core Ryzen, right?
Nope. Just averaging around 20%.
Here are the CPU stats from the last week.
But, I need 64GB of ram for my needs too!
No you don’t. I am using around 6GB / 16GB. The rest is utilized for cache. Of that 6GB, 4 of it is dedicated to the server 2019 VM. (Windows LOVES ram…) All of the above docker containers are running on only 2GB of ram.
As a bonus, my power consumption is quite low.
To note- 76 watts INCLUDES…
- 8 port unifi switch.
- 30 foot of RGB LEDs in one of my other bedrooms.
- 2x 5mp reolink RLC-520 poe cameras.
So- what was the point of this article?
Quite simply- to consider the actual requirements of your software you intend on running.
Once upon a time, I did run a plex server at my home, which utilized a older precision workstation, with a xeon, FB dimms, and 8x 2TB HDDs. Even on its quad core xeon, it could handle video encoding, and a few streams at the same time.
In the future, I intend on getting plex going on my server. Would the quad core ryzen cut it? More then likely. I will be rebuilding my entire gaming pc / server environment around mid-spring of next year… so, stay tuned.
My point is simple. You likely don’t need a big, expensive, or noisy server to power your services. Consider your needs, and your use-cases and go from there.
If you utilize a technology such as LXC or Docker Containers, it will GREATLY reduce your overall requirements, as you do not need to virtualize an entire operating system.
For storage- consider your requirements. If you have a need for storing 20+ terabytes of data, having hot-swap bays in an enterprise form-factor may be useful to you.
For my specific case- HDD failures are pretty rare. All of my 2TB drives are from the early part of last decade. They are still running strong. If one of the drives die, I do have both parity, and backups of all of my services. I might have to power my server down for 10 minutes to replace a drive… but, it’s really not a big issue.
In the end- it is your personal choice of which route you should take. Perhaps some of the enterprise features are very attractive to you. Perhaps you NEED IMPI. Or- perhaps a cheap workstation would fill your needs without issue. I am just asking you to consider your actual requirements!
I don’t think I made the intended point clear enough-
The intended point, is to evaluate your needs, and make an educated decision.
If you only want to run a minecraft server, an optiplex is going to work much better then a xeon with a slower clock.
If you want to run a huge plex server or NVR, It’s hard to argue against the low price, and ready power of a used server.
For my case, of a handful of home services, it would take 5 or so years before the cost savings of my 500$ server could match the low price of a decent rack mount server.
If your only goal is to run home assistant, and a few more tiny services, its hard to argue against a raspberry pi, with nearly non-existent power usage.
Quite simply- consider your requirements. Weight the pros and cons, and make an educated decision.
Just because somebody says, go buy a Dell R510, doesn’t mean its a good idea. Make sure it fits your size requirements, your noise requirements, and your power requirements.
Edit 3 –
I am literally not saying older hardware is bad, but, rather asking everybody to consider the variables, the needs, and the ROI.
As an example- lets say, you have a plex server you want to run, which theoretically will be loaded 100%, 24/7.
Lets say- you have a few options.
- An older server, such as the one at the top of this article. Lets pretend is uses 250w average, 24/7. It costs 100$.
- A much newer server, using a much more modern processor. Let’s pretend, it uses 150w on average, 24/7. It costs 300$
- A brand new server you built. Such as the one I built. Lets pretend, it uses 100w on average, 24/7. Lets say, it costs 600$
- A raspberry pi-4. Based on google, it uses around 4.2W. Lets pretend it does that on average, 24/7. (I know there is no way, it is going to make an effective plex server. Pretend for this example, that it can meet your needs)
I am going to use THIS calculator.
First- let’s calculate the power estimates from all three examples.
Now, lets calculate the energy from the 2nd server.
Lets put all of this data, into a table.
|Initial Cost||Energy Cost per Year||First Year Cost||Five Year Cost||Ten Year Cost|
|More Modern Server||300$||$157||$457||$1,085||$1,870|
|Brand New Server||600$||$105||$705||$1,125||$1,650|
Having a piece of consumer hardware running for 10 years, without maintenance is actually quite unrealistic. But- for this example we are going to assume you both, Use it for 10 years, AND, it never breaks.
Looking at the data- off the bat, the older server will much much cheaper initially. However- by the 5 year mark, the extra couple hundred spent on a more modern server, quickly pays off.
The brand new server, REALISTICALLY, will never never hit ROI, compared to the more modern server… because, lets face it. You are going to replace it long before 10 years.
Lastly- I added in a pi-4. Let’s assume it can run your workload without issues… well, there really isn’t a comparison. It essentially makes energy usage a non-factor.
The point of this article, once again, is to consider the variables.
If you spend 2,000$ on a piece of really fancy hardware, calculate the ROI versus say, a 200-400$ piece of used enterprise gear. If the optiplex linked above does the same job your 700$ server can do, how many years would it take to hit ROI? To note- I am NOT including the price of hard drives, and storage. This price will remain the same no matter how old or new your hardware is…. (Unless you buy some antique server which uses IDE drives…)