Short answer: Use three inputs, in this order, and each one has a published number behind it: your monthly budget, your target region, and how much system administration you will actually do. On the flat four-tier price list published by one vendor in this comparison set, the budget step alone narrows you to one of $10, $20, $40 or $80 per month, and the derived RAM-per-dollar ratio is exactly 0.100 GB per dollar on all four tiers - so paying more buys more capacity, not a better ratio. The region step does not change the price: six separately published location pages checked for this article all show the same $10 / $20 / $40 / $80 tiers, which makes location a latency decision rather than a price decision. The operations step is the one that actually disqualifies plans: official software requirements range from a 128 MB per-process floor to an 8 vCPU / 16 GB single-node baseline, and the official setup guidance for a self-hosted social server expects key-only SSH, fail2ban and a manually configured firewall. Below is the three-step matrix with the conclusion in the first screen, the full derivation of every cell, and an explicit list of what this matrix does not know.
The decision matrix, first screen
| Your three inputs | Conclusion from published numbers | Derived rule |
|---|---|---|
| Budget $10/month, one small service, comfortable with Linux | $10 tier: 1 vCPU / 1 GB / 20 GB NVMe | Budget band $10; workload floor 128 MB per process fits |
| Budget $10/month, but you need a full DevOps platform | No tier on this list qualifies | Published baseline 8 vCPU / 16 GB exceeds the 8 vCPU / 8 GB top tier |
| Budget $20–$40/month, several small services | $20 or $40 tier, sized by process count, not site count | RAM scales linearly with price; count your processes |
| Budget is hourly or the project is short-lived | Use a published hourly rate with a monthly cap instead | Hourly billing with a cap is a different contract, not a cheaper tier |
| Users are in Europe or Asia | Same price, pick the nearest published location | Six location pages show identical tier prices |
| You cannot configure SSH keys, a firewall and updates yourself | Do not start from an unmanaged tier | Official setup docs assume these tasks are yours |
| You need a control panel | Add its published license price to the tier price | Add-on is priced separately from the tier |
These conclusions are derived decision rules built for this article from published inputs, not recommendations quoted from any vendor. Each input is traced below.
One vendor publishes four tiers with flat monthly prices. [S1][S2]
| Tier | Monthly | vCPU | RAM | NVMe | Derived RAM GB per $ | Derived NVMe GB per $ |
|---|---|---|---|---|---|---|
| KVM1-US | $10 | 1 | 1 GB | 20 GB | 0.100 | 2.00 |
| KVM2-US | $20 | 2 | 2 GB | 25 GB | 0.100 | 1.25 |
| KVM4-US | $40 | 4 | 4 GB | 40 GB | 0.100 | 1.00 |
| KVM8-US | $80 | 8 | 8 GB | 80 GB | 0.100 | 1.00 |
RAM_GB_per_dollar = published_RAM_GB / published_monthly_price_USD
NVMe_GB_per_dollar = published_NVMe_GB / published_monthly_price_USD
KVM1-US: 1/10 = 0.100 GB per $ ; 20/10 = 2.00 GB per $
KVM8-US: 8/80 = 0.100 GB per $ ; 80/80 = 1.00 GB per $
All four divisions are derived arithmetic, not vendor quotes. The RAM ratio is flat, so the budget step is really a question of absolute capacity: how many gigabytes does your workload need. The NVMe ratio is not flat, so a storage-heavy workload gets more capacity per dollar on the cheapest tier than on the two most expensive ones. [S1][S2]
A second public price list shows a different shape of the same decision: an hourly rate with a monthly cap. One entry-level plan there is published at $0.00893 per hour with a $6 monthly cap, and a 2 GB plan at $0.01786 per hour with a $12 cap. [S3] That is a different contract from a flat $10 monthly tier, and the choice between them depends on how many hours the machine exists - a calculation that belongs to the cost-conversion comparison, not to this matrix. Here it is used only as the "hourly" branch of step one.
The vendor's homepage lists 18 cities under a heading that says 18 strategic locations, while its plan page says 17 locations and lists 18 cities, and its own body copy says 17 global locations. [S1][S2] The published counts disagree; this article does not resolve the discrepancy because no page explains it.
For this article, six individual location pages were read: Atlanta, Frankfurt, Singapore, Lagos, Marseille and Tokyo. [S4] Every one of the six shows the same $10 / $20 / $40 / $80 tier prices and the same separately priced control-panel add-on. [S4]
| Location page read | HTTP | Entry tier price | 2 GB tier price |
|---|---|---|---|
| Atlanta | 200 | $10 | $20 |
| Frankfurt | 200 | $10 | $20 |
| Singapore | 200 | $10 | $20 |
| Lagos | 200 | $10 | $20 |
| Marseille | 200 | $10 | $20 |
| Tokyo | 200 | $10 | $20 |
So on this particular list, changing region does not change the published price for the six locations checked. That makes region a free input: choose the location closest to your users. It does not mean latency is equal - no response-time or round-trip measurement is published on those pages, and none is estimated here. Six of the listed locations were checked; the remaining ones were not, so the table above is a sample, not a complete matrix of every location. [S4]
This is where a cheap tier stops being a budget question. The published requirements of four self-hosted applications span two orders of magnitude: [S5][S6][S7][S8]
| Workload | Published floor | What the operator must supply |
|---|---|---|
| File-sync server | 128 MB RAM per process minimum, 512 MB recommended per process [S5] | Process sizing and a web server |
| Forum platform | 1 GB RAM minimum with swap, 10 GB disk, Docker only officially supported [S6] | Docker, PostgreSQL, Redis, and swap configuration |
| DevOps platform | 8 vCPU / 16 GB single-node baseline, 40 GB for application nodes, swap discouraged [S7] | A machine larger than every tier on the four-tier list |
| Social server | No RAM or CPU number published; setup page expects key-only SSH, fail2ban and a firewall limited to SSH, HTTP and HTTPS [S8] | All of the above security setup, by hand |
The third row is the hard disqualifier. A published 16 GB single-node RAM baseline is exactly twice the top tier's 8 GB, and the same document says that for single-node installations the component storage requirements must be totalled on one machine and that its 40 GB figure covers application nodes. [S2][S7] Buying the $80 machine therefore does not satisfy that baseline; this article does not sum the remaining components because the plan page does not publish the component breakdown.
The fourth row is a different kind of boundary. Its official preparation page contains no hardware number at all, but it does state the operator's tasks explicitly, which is why "I can install a panel but not configure SSH and a firewall" is a legitimate matrix input rather than a vague skill question. [S8]
A control panel is a separate line item. The vendor's homepage prices a cPanel license starting at $26.99 while its plan page says from $26/month; the two figures differ, both are published, and neither is included in the $10 tier price. [S1][S2] A buyer who needs a panel must add it to the budget band in step one. The terms of service also list licenses among the items excluded from refunds. [S9]
Once you have checked a three-step plan decision, the VPS downgrade and rollback page shows which parts of an upgrade you can actually reverse.
While this page deals with a three-step plan decision, the cheap VPS entry price page separates the headline price from the price you are billed.
The figures here cover a three-step plan decision; for the part that follows, the cheap VPS for beginners page walks through deploying a first server in under an hour.
What this matrix does not contain: no measurement of latency, uptime, IOPS, throughput or support quality; no estimate of search volume, traffic or ranking outcome; no analytics data; and no reading of the two data sources that are reserved for the end of the 28-day reporting cycle. One vendor's pricing and legal pages returned HTTP 403 Managed Challenge on 2026-11-11, so that vendor contributes no new numbers to this matrix. [S10] The decision rules above are constructions of this article, derived only from the published inputs listed under Sources; they are not vendor claims.
A buyer with a $20 monthly budget, users near Singapore and one small file-sync service:
step 1 budget $20 -> KVM2-US (2 vCPU / 2 GB / 25 GB NVMe) [S1][S2]
derived RAM per $ = 2 / 20 = 0.100 GB per $
step 2 region Singapore -> Singapore location page, same $10/$20/$40/$80 tiers [S4]
price delta vs Atlanta = $20 - $20 = $0.00
step 3 workload floor = 512 MB per process recommended [S5]
derived upper bound on processes = floor(2048 / 512) = 4
(ceiling only; OS, web server and database are not included,
and their footprint is not published, so the true number is unknown)
conclusion the $20 tier is the smallest tier that satisfies this
published floor; add any control-panel license separately [S1][S2]
Nothing in that example is a measured result. The "4 processes" figure is a ceiling computed from two published numbers, and the article explicitly says the real capacity is unknown.
Start with the three inputs in order. Budget fixes the tier, region is a latency choice that does not change the published price on the six location pages checked here, and operations capability decides whether an unmanaged machine is appropriate at all. The first-screen table gives the conclusion for each combination.
On the six location pages read for this article, no: all six show the same $10 / $20 / $40 / $80 tiers. That is a price observation, not a latency measurement, and the remaining locations were not checked.
Because the inputs are yours, not the vendor's. A published 16 GB baseline rules out every tier on this list for one workload, while a 128 MB per-process floor fits the cheapest tier for another. A matrix returns a conclusion per combination; a ranking would require measured performance data that none of these pages publishes.
All pages read on 2026-11-11.
No sentence in this article is reproduced from the three compared sites; only published prices, plan parameters, policy terms and official requirement figures are used. Every ratio and process count is labelled derived arithmetic or a computed ceiling, no measured performance figure appears, and no analytics or ranking data is used.