All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Desired memory for low end VPSs
slowservers
Member, Host Rep
Hi LET,
Host nodes for Slow Servers are definitely on the small side, and they are a bit unusual. This manifests in plans and pricing.
I was wondering about making a fairly small adjustment to my node configuration and if it's something that would actually interest customers or not.
I designed the hosts to use as little power as possible, to be built off old hardware, and to work well with my choice of hypervisor (OpenBSD.) This is a bit of an unusual set of criterea.
This a reference node looks like this:
CPU: Opteron 4162EE (6 core, 1.7GHz)
Memory: 32GiB ECC
Storage: 1TB Samsung 850 PRO or 1TB Velicoraptor
This gives me multiples of 1GB of memory per 32GB of storage, which is certainly above average. I don't do the whole alacarte CPU/memory/disk -- it doesn't work well with small hosts.
One major limitation is that OpenBSD's VMM can only issue one vCPU, so anything genuinely CPU heavy is out. Another limitation is that OpenBSD's scheduling isn't great, and if every core is pegged, performance drops rapidly. So keeping a fairly generous core to VPS ratio is important.
While I've seen Xen run and schedule 256MB VPSs admirably (with 4 vCPUs, what the host had!), it just isn't OpenBSD's forte at the moment. And OpenBSD itself, if a guest, will swap much less than 768MB just on kernel relinking. So this puts me in an unusual category of low power/performance, but not capable of lots of very low power servers (think Tierhive.)
Ultimately, my pricing is simple, but high: $6/GB of memory, per 32GB disk. $1.50/month extra for IPv4.
What I realized is that for a small increase in power use, I could double the host memory with registered dimms (so 64GB per node instead of 32GB.) This wouldn't be that expensive up front. I'd have to double check power usage, but I'm guessing it would add a few watts. Now with this, I wouldn't want to host more VPSs. I would just offer more memory for roughly the same amount of money, basically.
The current smallest size is 768MB. So I might end up with a larger minimum size of 1.5GB of memory. Unless I increased disk space, it would then be 1GB memory/16GB disk.
So the smallest $4.50/month plan would get bumped up from 768MB to 1.5GB memory with 24GB of disk.
On the larger end of things, it becomes less useful -- when are you going to use more than 16GB of memory with one vCPU? I do currently use a 16GB / 512GB disk / 1 vCPU server succeessfully as a Monero now (on a Velociraptor, nonetheless!) Doubling that memory to 32GB seems useless in this application.
I could potentially add an extra drive at the same time, also increasing power usage but not tremendously. At that point I don't think I'd give twice as much for the same price -- I'd have to evaluate it.
I know a lot of people are happy with 512MB memory servers. Some are happy with 1GB. And some want more. But the ones that want more, do you also need more CPU to go with it? Or would this make my offering more usable?
Thank you!
PS: Here's my current reference node: https://slowservers.net/hosts/loki/
And some ramblings on pricing, if you just can't get enough of this long form content: https://slowservers.net/gemlog/2026-04-25-on-pricing.gmi
- Would you be considerably more interested in this server plan or another?23 votes
- 1 vCPU / 768MB memory / 24GB disk for $4.50/month17.39%
- 1 vCPU / 1.5GB memory / 24GB disk for $4.50/month43.48%
- 1 vCPU / 1.5GB memory / 48GB disk for $5.00/month39.13%
- Would you be considerably more interested in this server plan or another?23 votes
- 1 vCPU / 16GB memory / 512GB disk: $96/month8.70%
- 1 vCPU / 32GB memory / 512GB disk: $96/month4.35%
- 1 vCPU / 16GB memory / 256GB disk: $48/month13.04%
- 1 vCPU / 16GB memory / 512GB disk: $48/month47.83%
- 1 vCPU / 32GB memory / 1TB disk: $96/month26.09%
- Should I double memory capacity?23 votes
- Yes, it's worth a few more watts73.91%
- No, save every watt!26.09%
- Should I double disk capacity and memory?23 votes
- Yes, this would influence my buying decisions.56.52%
- No, it's not worth the power.4.35%
- Slow Servers is stupid.39.13%
- How much would you pay for 1 vCPU / 1.5GB / 24GB disk? ($1.50/month extra for IPv4)23 votes
- $7/year60.87%
- $3/month8.70%
- $4/month13.04%
- $5/month4.35%
- $6/month0.00%
- $7/month4.35%
- These are all too expensive!8.70%
- What is the smallest useful server memory for you?23 votes
- 256MB26.09%
- 512MB13.04%
- 768MB4.35%
- 1GB26.09%
- 1.5GB0.00%
- 2GB17.39%
- Bigger..13.04%
Comments
I was looking at your services the other day
I really like the idea and goal, but the 1vCPU limit seems to be the only killer.
Why not use something like Proxmox that's still free, open source, etc but can handle at least assigning multiple vCPU? Or look for a patch for your software? I saw in the website that it says the issue is your software you use.. but that's probably a HUGE issue for clients.
The old hareware recycling is GREAT. I really love this idea and it's awesome. But -- if you offer plans with huge RAM and Disk, the CPU cores need to match up somehow to make it work. Otherwise, the RAM becomes hard to even use what you paid for. If that makes sense?
Anyways - I really do love your project
Much love
8 MB RAM for 1$ year when?
I wouldn't consider buying any vps under 1GB of ram. 1GB is minimum.
Having 16GB with 1 vCPU just doesn't make any sense considering you can get more vCPU from other providers.
Also the pricing per month is just too expensive.
With such a low end node, you are much better off selling the whole node like NOCIX/WholesaleInternet
I think a VPS should have at least 64 MB of RAM, at least 512 MB of storage, cost no more than $7 per year, use KVM virtualization, and include NAT IPv4 + IPv6.
Any plans to also offer Shared Webhosting one day?
I think it's difficult to run more than a couple services with at least 512MB & I wouldn't consider any server with less. Would rather have more servers on the very cheap end of the price range than on the very high end (this is LowEndTalk, after all).
Thank you! I appreciate that!
OpenBSD was an idealistic choice, mostly. There's some things about it that are easier and maybe better for VPS hosting. I just like OpenBSD a lot in general, and working with something that I like is nice.
I think virtual SMP (more than one vCPU) support will come eventually, just not sure when. Disk performance is a major issue on OpenBSD as well. It's usable, just not competitive.
I've used Xen before and would probably use it, if not OpenBSD. I might consider it on some different hardware. Of course if a couple things could be remedied on OpenBSD, it would make it a lot more practical.
I might actually be expanding quite a bit with some newer hardware being generously offered for free, so for that buildout I might do things differently -- we will see. I still have to get it, sort through it, and test it out.
Yeah, that's understandable. It does work for Monero nodes, it's just not performant.
I understand that. I'm used to selling in the KYC-free space where prices are a lot higher.
That would be interesting to sell dedicated servers. I'd have to really work on my networking setup to use vlans and have sufficient filtering at the router level. I wonder how much demand there would be.
Sounds like a great option through Tierhive!
Actually, I would like to! Probably very basic, static only hosting though.
That's what I was thinking, that most of you would rather have more cheaper servers than a few bigger servers. I used to lean towards the many-small camp, now I think I prefer a bigger server (or two) per point of presence. Of course as small as can handle the load is nice as well.
1vCPU 512MB RAM 3GB SSD KVM virtual technology that's awesome
It’s a tough sell because there are plenty of KYC free providers here and many are expecting lowest prices and the best deals here on LET.
Thank you to everyone for your feedback!
I am lowering prices and will be adjusting hosts for double memory+disk capacity. It turns out that 32GB to 64GB of memory uses about 3 more watts. And going from a 1TB to a 2TB SSD should use a negligible increase in power.
For the curious: https://slowservers.net/gemlog/2026-06-06-price-reduction.gmi
(Or on Gemini: gemini://slowservers.net/gemlog/2026-06-06-price-reduction.gmi )
It's still not cheap, but it is cheaper.
Depends entirely on the workload, but my rough rule: 1GB is fine for a single service (small site, bot, DNS), 2GB once you add a database, 4GB+ if you're running Docker with a few containers. Swap helps for occasional spikes but you don't want to lean on it. Honestly RAM is the spec I'd upgrade before CPU for most low-end use.
Yes, give us those **real ** lowendVPS back: 256MB RAM, 3GB disk, 1vCPU and IPv4/IPv6 for 5 USD per year. Missing those old discussions from +10 years ago when we were optimizing debian for prividing DNS, nginx/php etc.
My needs have changed over the years, but I'm generally not interested in anything with less than 2 GiB memory these days.
After I've installed Elastic Agent, Nessus Agent and my RMM, I'm losing 300/400MB anyway.
While it might not be overly practical i'll give you a thought to play with that (i think) would kind of fit your business model: Try to acquire a bunch of solar panels/batteries so you have a supply of cheap energy, get some relatively fat fiber line (something like a gbit - price probably depends a lot on your location but maybe it's feasible?), connect a bunch of cool retro servers to your cheap energy and make them available to the world over some GRE tunnel or similar (to avoid DDoS problems on your uplink - not exactly free but also shouldn't totally kill your budget). Voila you've build a reasonably functional shed DC.
Obviously getting enough panels and batteries would be a noticeable investment and probably makes the whole thing uneconomical unless you can get some kind of killer bargain but in general i think it would fit the overall vibe nicely and maybe allow you to get a bunch of competitive deals online. Obviously the setup has a couple downsides (mainly reliability wise and to a certain degree available bandwidth - 100mbit ports aren't exactly great but would still allow 10 boxes on your line before you even oversell anything) but i'm pretty certain people would look beyond that if the price is right.
Something like a dual 2637v2 is still a nice box if you aren't paying much for it and such a config with 16-32GB RAM and a bunch of smallish disks for ghetto raid shouldn't really cost much more than (maybe) $100 to build (could probably be cheaper given a bunch of good catches on parts). With a tiny bit more of an investment (like 2643, 2673, 2667, ...) the resulting box could actually even still be somewhat impressive.
IPv4 for $5/year is really pushing it, but I see what you mean. That would be a lot more possible if I were using Xen. Arbitrarily, I chose OpenBSD, which I'm not as comfortable loading up after my testing. It just doesn't schedule as nicely.
I think that's already done well with TierHive, though. Some great discussions on LES about tiny server configurations.
I was looking into hosting hardware on my property and wanted to have actually off-grid (minus the fiber) power. I had an Atom-based configuration, initially, but VMM did not work on it. It would've been really cool, and amazingly low power, but missing the right processor extension. Also not much of that hardware available, sadly.
Unfortunately, I couldn't get fiber here. I had one opportunity to host in the nearest town that would've more of a shed situation (maybe with some solar potential) but it never panned out in the end.
Now this area isn't that great for solar. It's doable, but going to need a lot more panels than some areas. On the bright side, cooling needs are lower, so that can help.
Sir its 2027, lowend has a new standard of lowend since almost everything include people's income has been rising over the past 10+ years (as you missing).
What you trying to mention here is poverty-end, not lowend anymore.
I thought people love $5/year
Yeah, that's pretty much the crux. Basically all retro configurations that still pack a (relative) punch are going to be energy hungry monsters. In that respect investing a couple $ more into a bunch 2643v2, 2673v2, 2667v2 and so on is probably going to be more economical in the end since they bring a lot more power than the literal $5 2637v2 while having basically the same TDP.
If low power is more of your thing though you might take a look into thinclients. There's a bunch of quite modernish one's floating around for cheap by now. Not too long ago i saw a a stack of HP t740 (Ryzen Embedded V1756B - https://parkytowers.me.uk/thin/hp/t740/) on Ebay for 100€/unit shipping included. Sure the embedded Ryzen might not be that energy efficient but for its ~50W TDP its also doing pretty well (outside of having integrated graphics that might actually not be half bad) compared to a lot of (assuming single CPU config small) E5s (https://cpubenchmark.net/cpu.php?id=3574&cpu=AMD+Ryzen+Embedded+V1756B vs. for example https://cpubenchmark.net/cpu.php?cpu=Intel+Xeon+E5-1620+v2+%40+3.70GHz&id=2047 / https://cpubenchmark.net/cpu.php?cpu=Intel+Xeon+E5-2637+v2+%40+3.50GHz&id=2178 and 2637 is one of the few dual socket v2 E5s that actually has somewhat tolerable singlecore performance after all) and there's likely models with lower ratings too.
Yeah, i already figured that it was a bit of a pipe dream but who knows maybe at some point there might be an opportunity making it a bit less of a pipe dream
I figure it depends a lot on where you are located but at least around here you are one lucky bitch if your increase in income is able to offset inflation. Actually the average income raise probably hasn't been sufficient to cover inflation in like the last ~40 years, so even if people see bigger numbers at the end of the month they are actually being paid less.
Realtalk and my curiosity: Opteron 4162EE who in their right mind would pay 5$ per month for this piece of shit?
I'm not getting paid, so I can't say much. T_T
You've definitely thought about this! I was curious and decided to compare it to my Opteron 4162EE.
https://www.cpubenchmark.net/compare/2178vs2477/Intel-Xeon-E5-2637-v2-vs-AMD-Opteron-4162-EE
There's clearly no comparison there, it just blows it out of the water. 130W vs 35W TDP, though. The Opteron does have 6 cores, vs 4.
VMM only supports single core VMs, as it is, and I'm specifically trying to take a "greener" angle on it. Slow Servers is just not a good option for anyone wanting any notable performance. There's so many hosts that do that better and I'm not trying to compete with them.
I have a metered PDU. Currently have four servers running, two routers, and a switch. The PDU is reading 2 amps, and that's with a 120V feed.
I did end up with a couple of higher TDP Intel hosts as part of my testing.
https://slowservers.net/hosts/ullr/
https://slowservers.net/hosts/prometheus/
What I found is that I could get impressively low power draw with setperf=1, even setperf=30, but at modest workloads the Opteron was more efficient. If I let the Xeon run unthrottled with setperf=100, it would be good at idle but spike up a lot under load. I just felt the most comfortable with the relatively consistent power use on the 4162, while having consistent performance.
There's one other draw towards AMD over Intel, and that's trustworthiness. I'm not going to say that AMD is entirely trustworthy, as far as backdoors or things like Intel ME go, but in general AMD has a better reputation. One my first customers was excited to switch over to AMD as soon as I offered it. Now some Intel gear is pre-ME, and probably fine. And maybe all Intel gear with ME is fine. But I can understand having concerns about having a full Minix kernel running next to everything else.
Certainly not the only supply chain security concern, and not one that everyone will have, but it's another thing that tips the scales for me, at least for this deployment.
That is also faster than the 4162! There may well be other configurations that make more sense. Can you get those with ECC?
Like my Atoms of choice, which I've pretty much bought up all of the affordable stock of on Ebay (I have 5x Atom S1260 motherboards in total.) I've kind of pigeonholed myself with these Opterons. Basically, there's one single socket motherboard that works out well for my 4162, and I've bought about as many as I can. I figure I'm committed to having a fleet of ten 4162 hosts, plus a spare, maxing out the 1/3 cab. I did see a couple of Dell 2U hosts that could take two 4162s, and that might be another possibility. I do prefer spreading out failure risk whenever possible though, and would rather have more distinct servers hosting the same number of servers. This applies both to actual failure, reboots (waiting for fscks of guests), and noisy neighbors.
Planning on phasing out Ullr and Prometheus at some point, in favor of just the 4162 chassis.
If I get to the point of thinking about expanding, I'll probably have to reevaluate my choice in hardware, and it might be a good idea anyway.
That would be cool, for sure! There is one other concern that I've had. If I can get fiber to a rural place, it's almost certainly going to be a single point of failure and not in a loop. And regardless, it will probably be one carrier without as good of routes, and one place to lose service. A fiber cut might not happen often, but I could easily see having a fiber cut every few years, with at least a few hours of downtime. Maybe not the biggest deal in the grand scheme of things.
You're right, there's tons of other options out there. For the exact same cost, or often less, you can get faster servers elsewhere.
A lot of people don't need the performance, though. I fully believe in dogfooding, so my servers of importance are hosted on these hosts. I have a couple of servers hosting Gemini + HTTPS sites, that also have Prosody running for XMPP. The hosts could be half as slow as they are, and I probably wouldn't notice.
Now where I do notice is the SporeStack API. SporeStack runs on a 2GB Slow Servers instance. Going from Vultr to Slow Servers was about a 10x performance hit. (It's on Prometheus, which is a E5-2650 v2 but at setperf=1, so 1.2GHz. It also has a HDD and not a SSD. It's ultimately even slower than the 4162-based host(s.))
Part of it may have been related to switching from Debian to OpenBSD (guest side,) as well. But at least 5x slower, no question. I did end up refactoring a few things to make it more efficient. And in practice, it hasn't been terrible. I have one user who pushes that API pretty hard and I made some changes to accomodate (and largely to force the user into doing sequential launches.) I don't think I've lost sales and I haven't had anyone complain about the performance, other than that one user. I would like to improve it, and there's a few things I plan on doing that should make it better. But it's been okay as it is.
Anyway, this is getting pretty long-winded, but the core of my business that we live off of runs on a $8/month server and it's fine. Great? No, but good enough. That same server was $10/month at Vultr (or maybe $12/month, depending on the lineup it was from.)
And I get it if it's too slow. I just know that many can find a lower level of performance adequate, while keeping more hardware out of landfills. That's more important to me than having things run faster.
and usually there would be 2 off those bad boys on the board. Without some seriously cheap/free energy it's probably going to get out of hand fast
Sure, i mean, it's not like you can't run anything on lower specs. What's going to affect you most (outside of magical free energy Christmas land) is the performance to watts/space ratio or basically how much clients you can support with how much watts/space as that's what in the end is going to dictate your prices. Maybe you could gain a bit of headroom looking into lighter (something like LXC, jails, ...) or even no (shells) virtualization at all since i think there's a fair chance clients who don't need much computing power might also not need a full system.
Oh, that's a pretty interesting idea. Sadly most (pretty much all?) server boards don't really allow for a lot of fine tuning but as long as you're fine with more enthusiast type boards (i vaguely remember there even being a bunch of Chinese dual socket Xeon boards aimed at a gaming audience) you could probably even push this one step further and go into underclocking/undervolting. There might actually even be bit of uncharted territory there as most tinkerers usually stuck to overclocking but going the opposite direction still occasionally managed to get some crazy results in terms of efficiency.
As far as shadow systems is concerned AMD sadly isn't far behind Intel. I don't remember what AMD's version of ME is called but if i'm not fully mistaken they closed the gap pretty quickly (last non-ME-and-friends Intel CPU is the Core2 Duo, for AMD it would probably be the Phenom - i'm not really sure how that maps to the various server/embedded CPUs though).
Well, if it sells it sells
100%. Going pre-ME is still sadly pretty rough though.
That's a good question to which i sadly don't really have an answer for. I guess it's more likely to a no though. I mean those things are usually intended to be glorified terminals, so adding ECC might not make a lot of sense to the intended audience. If the boards would take ECC though (assuming the CPU supports it)... unless someone has tried and posted about it (outside of parkytowers information is usually very, very sparse though - if you haven't checked the link you might want to as there's a ton of info on the hardware of various thinclients that pass through Ebay each day for next to nothing) there's probably nobody besides the designer of whatever reference this was based on and maybe HP who would know.
If the CPU supports it one could obviously look for offers on (actually documented) standalone boards but from my experience those tend to be expensive (lowish volume production aimed at industry) and the t740 board is already pretty nice (at least as far as thinclients go - so many expansion options including an actual PCIe slot too, which isn't that common with those boards) and compact on top of that (for comparison the Igel H860C has a somewhat similar configuaration but hardly any expansion possibilities outside of USB: https://parkytowers.me.uk/thin/Igel/ud/ud7/H860C/). Personally i'd probably try to mcgyer some multi-board PSU and enjoy the cheap supply
Yeah, given you can likely pack multiple Atoms into the space they are taking up i guess that makes sense. The L Xeons are in a somewhat weird spot anyways.
Obviously reliability wouldn't exactly be a strong point. There's no denying that but like i said i'm pretty sure there's enough people coming from a lowend angle to look past that. I don't know if you have been lurking back then (if you did i'm sorry for telling you old news) but something like this has actually been attempted around here a couple years back by a Romanian guy. I'm certainly not saying that there wasn't any problems but in the end i guess it was mostly the fact that he hadn't factored in the possibility of his line being DDoS'd.
Sure, hiding behind a tunnel isn't all that economical when it comes to low cost VPS but on actual dedis something like $2-3/m for an IP and a TB of free traffic probably wouldn't be that bad (with a bit of scale i figure there could be better deals - this is just more or less what you'd get out of the box at noez or other tunnel providers).