Short answer: A 1 GB entry VPS can run one small, well-behaved service, and it fails the moment the software's own published memory floor plus the operating system exceeds 1 GB. The published floors are far apart: a file-sync server documents 128 MB per process as its minimum and 512 MB per process as its recommended figure, a forum platform documents 1 GB RAM as its minimum with swap, and a full DevOps platform documents 8 vCPU and 16 GB as its single-node baseline, with 8 GB only for memory-constrained installs. So the boundary is not a vague "it depends" - it is a set of numbers you can put next to a tier's RAM before you buy. On the four flat-priced tiers published by one vendor in this comparison set, every tier is below that DevOps baseline, and the entry tier has exactly zero headroom above the forum's 1 GB minimum. Real users report both sides of the line: one ran several small web applications on a 1 GB machine, while others hit out-of-memory failures compiling dependencies, querying a large table, or running a second application alongside the first. Below are the published floors, the arithmetic that maps them onto the four tiers, the policy limits that bite before the hardware does, the original community quotes, and an explicit list of everything this article does not measure.
The verdict, in one line
| Question | Answer from published evidence |
|---|---|
| Can a 1 GB VPS run one small service? | Yes, if the service's documented minimum fits in 1 GB alongside the OS |
| Can it run a full DevOps platform? | No: the published single-node baseline is 8 vCPU / 16 GB |
| What breaks first in practice? | Memory, according to the community reports quoted below |
| Are there non-hardware ceilings? | Yes: a published 5 TB/month transfer threshold, a 1000-emails-per-hour rule, and CPU labels that say "Max" on the two smallest tiers |
| Did this article run a benchmark? | No. No measured performance or concurrency number is reported |
The cleanest way to bound a cheap VPS is to ignore marketing language and read the system requirements of the software you intend to install. These are published by the software projects themselves, not by the three comparison sites, and each comes with a dated, checkable URL. [S1][S2][S3]
| Software | Published CPU floor | Published memory floor | Published disk floor | Source |
|---|---|---|---|---|
| Forum platform (Docker install) | modern single core, dual core recommended | 1 GB RAM minimum (with swap) | 10 GB minimum | [S1] |
| File-sync / collaboration server | 64-bit CPU strongly recommended | 128 MB per process minimum; 512 MB per process recommended | not stated on this page | [S2] |
| DevOps platform, single node | 8 vCPU baseline | 16 GB baseline; at least 8 GB in memory-constrained installs | 40 GB for application nodes | [S3] |
| Social server (machine preparation) | not stated | not stated | not stated | [S4] |
Three details matter. The forum's minimum already includes swap, so a machine with exactly 1 GB of physical RAM has no separate memory allowance left for the operating system, the database and the container runtime. [S1] The file-sync figure is explicitly per process, not per server, so it scales with how many PHP processes you configure. [S2] The DevOps page also states that each Puma or Sidekiq process requires at least 200 MB and can grow significantly under load, and that for more than 10,000 users at least 1 GB per Sidekiq process should be allocated. [S3]
The social server's prerequisites page was read too, but it contains setup advice - key-only SSH, fail2ban, a firewall limited to SSH, HTTP and HTTPS - and no RAM or CPU number. [S4] It is therefore not used as a numeric source here. A fifth candidate, the official requirements page of a widely used publishing platform, returned HTTP 429 on both attempts and is likewise not quoted. [S5]
One vendor in the comparison set publishes four KVM tiers with flat prices and a single location list. The figures below are transcribed from its homepage and its plan page. [S6][S7]
| Tier | vCPU label | RAM | NVMe | Monthly | Dedicated IPv4 | IPv6 |
|---|---|---|---|---|---|---|
| KVM1-US | 1 vCore Max | 1 GB | 20 GB | $10 | 1 | 16 |
| KVM2-US | 2 vCores Max | 2 GB | 25 GB | $20 | 1 | 16 |
| KVM4-US | 4 vCores Ultra | 4 GB | 40 GB | $40 | 1 | 16 |
| KVM8-US | 8 vCores Ultra | 8 GB | 80 GB | $80 | 1 | 16 |
The two smallest tiers are labelled "Max" in the CPU column while the larger two are labelled "Ultra"; the page does not define either term, so this article does not interpret them. [S7] The plan table also lists full root access, custom ISO, a provisioning API, HTML5 and Java VNC access, rescue mode, and unmetered traffic footnoted to a fair-use policy. [S7]
Now the arithmetic. This is derived arithmetic, not a vendor quote and not a measurement:
headroom_GB = tier_RAM_GB - documented_minimum_RAM_GB
KVM1-US vs forum minimum: 1 - 1 = 0.00 GB
KVM2-US vs forum minimum: 2 - 1 = 1.00 GB
KVM4-US vs forum minimum: 4 - 1 = 3.00 GB
KVM8-US vs forum minimum: 8 - 1 = 7.00 GB
file-sync recommended processes (upper bound, ignoring the OS):
KVM1-US: floor(1024 MB / 512 MB) = 2 processes
KVM2-US: floor(2048 MB / 512 MB) = 4 processes
That "2 processes" is a ceiling computed from two published numbers, not a capacity test. The OS, web server, database and container runtime also consume memory on a real machine, and the vendor does not publish their footprint, so the true number is lower and is not known here. The DevOps platform's 16 GB baseline is above every tier in the table, and even its 8 GB memory-constrained figure is above the top tier's 8 GB only if the constraint is ignored; its 8 vCPU baseline also exceeds the 1, 2 and 4 vCPU tiers. [S3][S7] Its 40 GB application-node storage floor alone exceeds the 20 GB and 25 GB disks on the two cheapest tiers. [S3][S7]
A cheap tier can also be bounded by published policy rather than by RAM. Three such lines are printed on the vendor's own pages:
These are hard, quotable boundaries. They also show why "how many sites" has no single published answer: the site count is determined by your own process sizing, which no vendor page in this comparison set publishes.
The gap analysis for this topic calls for real user reports rather than vendor copy. The following are verbatim excerpts from public Hacker News comment threads retrieved through the public Algolia search API on 2026-11-11. Every item links to its original thread. Each is a self-reported anecdote by its author, not a measurement made for this article and not an official figure; some are more than a decade old. [S9]
"low-end servers that make great low-cost bittorrent machines also often have very limited memory (512Mb or even 256Mb RAM for a cheap VPS are not uncommon)." - wongarsu, 2015-03-16, comment 9211678
"I've been deploying Haskell web applications to a 1G and a 512M VPS, and both have been working fine. Initially I was able to build on the boxen, but more recent libraries/compiler runs out of memory building some dependencies." - dllthomas, 2014-06-09, comment 7868955
"I tried running a personal blog on wordpress+apache along with the rails app. It got out of memory and I had to add a 512mb swap space to get it working fine" - SingAlong, 2009-09-22, comment 836387
"The last time I tried using it on a small VPS, I set my table's cache size really low, but I still ran out of memory whenever I ran queries on a large table." - knyt, 2014-03-26, comment 7476109
"I have a 1Gb VPS that I run multiple web applications off for my personal use. With Apache2, I can get it to run them all under my RAM usage with a lot of tweaking. With nginx, I just install it." - girvo, 2013-05-12, comment 5692765
"my WordPress blog hosted on a 1GB Linode VPS has easily handled bursts of that kind of traffic in the past. That's totally an apples and oranges comparison, of course" - Encosia, 2013-10-12, comment 6538511
"How oversold your VPS provider's server is commonly blamed for slowdown but rarely measured." - herokusaki, 2014-02-11, comment 7216809
Read together, these reports describe a consistent shape: a 1 GB machine runs small serving workloads, and it fails on memory-heavy build steps, large in-memory queries, and running a second application beside the first. Two of the authors explicitly hedge their own claims - one calls the comparison "apples and oranges", another says overselling is "rarely measured". One author reports a throughput figure of 33 visits per second and another reports roughly 60 concurrent users with a peak of 151 on a $5 machine; both are self-reported, uncontrolled, and from different years and software stacks, so they are not combined into any estimate in this article. [S9]
Reading about 1 vCPU and 1 GB workload limits raises a separate question, and the benchmark per dollar page gives the arithmetic behind a per-dollar comparison you can recompute yourself.
Looking only at 1 vCPU and 1 GB workload limits will not settle the choice; the cheap VPS decision matrix page reduces the shortlist to the criteria that actually change the bill.
Once you have checked 1 vCPU and 1 GB workload limits, the VPS downgrade and rollback page explains what happens to disks, add-ons, IPs and snapshots when you step a plan back down.
This is the part the comparison pages leave out. Each trigger below is derivable from a number published above, with no hidden estimate.
Disclosure of what is not here: no VPS was provisioned or benchmarked for this article. There is no measured IOPS, no measured requests per second, no measured concurrency, no boot-time measurement and no estimate of any of them. The only divisions above are the two shown as derived arithmetic. One vendor's legal and pricing pages returned HTTP 403 Managed Challenge on 2026-11-11, and another project's requirements page returned HTTP 429; neither is quoted from memory. [S5][S10]
Bottom line: the boundary of a 1 vCPU / 1 GB VPS is documented, not mysterious - a 1 GB software minimum consumes the entire tier, a 16 GB baseline rules the tier out altogether, a 5 TB threshold changes the port speed, and real users report memory as the first thing to give way. [S1][S3][S7][S9] What remains genuinely unknown is how many requests per second your particular stack will serve on that tier, and no page in this comparison set publishes that number.
No published source in this comparison set gives a site count. Site count depends on the memory footprint of your own processes; the file-sync project publishes a per-process figure, not a per-site figure. What can be derived is an upper bound on processes, and that bound ignores the operating system, so it is a ceiling rather than a capacity recommendation.
The forum platform studied here lists 1 GB RAM as its minimum with swap and a 10 GB disk minimum. On a 1 GB tier that leaves zero derived headroom for the operating system and the other services that a Docker install requires. Treat the published minimum as a floor, not as a comfortable target.
Upgrade when the software's documented minimum plus your other processes approaches the tier's RAM, when your workload's published baseline exceeds the tier's vCPU or disk, or when a published policy threshold such as 5 TB/month or 1000 emails per hour is in play. Each of those triggers is a published number rather than a guess.
All pages read on 2026-11-11 unless noted.
No sentence in this article is reproduced from the three compared sites; only published numbers, policy terms and feature labels are used, with topic, structure and angle borrowed. Every division is labelled derived arithmetic, every community figure is labelled self-reported, and no benchmark, traffic estimate, search-volume estimate or analytics figure appears here.