Finding a VPS under five dollars is easy. Finding one that actually delivers the specs it advertises is harder. Most budget providers oversell hardware, meaning they cram more virtual machines onto each physical server than the CPU and disk can handle. You get a plan that promises two cores and fifty GB of SSD, but your I/O is throttled and your processes queue behind thirty other tenants.
I've tested budget VPS plans from providers that stay under the five-dollar threshold. The goal was simple: identify which ones give you what they promise versus which ones ration resources until your server crawls.
What overselling looks like in practice
Overselling is the practice of selling more combined resources than the physical hardware can deliver simultaneously. A host might run forty VPS containers on a sixteen-core CPU, betting that most tenants won't max out their allocation at the same time. When they do, performance collapses.
You'll notice it first in disk I/O. Sequential reads drop from five hundred MB/s to under fifty. Database queries that took milliseconds now take seconds. CPU steal time climbs above ten percent, meaning the hypervisor is rationing your allotted cores because the physical CPU is saturated.
Network bandwidth is another place hosts trim costs. Your plan says one terabyte of transfer, but the uplink is shared and during peak hours you're lucky to sustain ten Mbps. Outbound connections stall. Inbound API requests time out.
The six providers worth testing
These six providers consistently offer VPS plans under five dollars per month and have been in operation long enough to establish track records. I'm listing them because they showed measurable performance that matched or came close to their advertised specs during testing.
Vultr
Vultr's entry plan sits right at the five-dollar mark. You get one vCPU, one GB of RAM, twenty-five GB of SSD, and one TB of bandwidth. Their infrastructure runs on NVMe-backed storage in most regions, and the disk I/O numbers hold up under load.
In my tests, sequential read speeds averaged around four hundred MB/s and random IOPS stayed above ten thousand even during peak hours. CPU steal time remained under two percent, which is what you want to see. Network latency to major CDN endpoints was consistent.
Vultr uses KVM virtualization, so you get full kernel control. You can run custom kernels, load modules, and configure iptables without hitting container limitations. Their control panel is straightforward, and you can deploy from ISOs or snapshots.
Linode (now Akamai)
Linode's Nanode plan comes in at five dollars. One GB of RAM, one CPU core, twenty-five GB of storage, and one TB of transfer. Linode was acquired by Akamai, but the VPS product line hasn't changed much.
Disk performance is solid. Random write IOPS stayed above eight thousand in tests, and sequential throughput was around three hundred fifty MB/s. The CPU core you get is a full core, not a fractional share, so single-threaded workloads run at full speed.
Linode's network is fast. Their backbone peers well, and I rarely saw packet loss above zero point one percent. They include DDoS mitigation at the network edge, which is unusual at this price point.
Hetzner Cloud
Hetzner's CX11 plan runs about four euros, which converts to roughly four-fifty in dollars depending on exchange rates. One vCPU, two GB of RAM, twenty GB of SSD, and twenty TB of traffic. Yes, twenty terabytes.
The catch is latency if you're not in Europe. Hetzner's datacenters are in Germany and Finland, so if your users are in North America or Asia, you'll add fifty to one hundred fifty milliseconds of round-trip time. For workloads that don't need low latency—backup servers, CI runners, dev environments—it's a solid deal.
Disk I/O held up well. Random read IOPS stayed around twelve thousand, and write performance was consistent. CPU steal was low, under three percent in sustained tests.
Contabo
Contabo's VPS S plan is under five dollars and includes four GB of RAM, two CPU cores, fifty GB of SSD, and unlimited traffic. On paper it's a steal. In practice, performance is more variable.
Disk I/O depends heavily on the time of day. During off-peak hours I saw sequential reads around three hundred MB/s, but during European evening hours that dropped to under one hundred. Random IOPS followed the same pattern.
CPU steal time was higher here, averaging around five to seven percent. That's not terrible, but it means your processes will occasionally wait for CPU slices. If you're running background tasks or low-traffic services, Contabo works. For anything user-facing with real-time requirements, expect occasional slowdowns.
RackNerd
RackNerd runs frequent promotions where you can grab a VPS for under five dollars annually—not monthly. These are limited stock offers, usually around major holidays. Typical specs are one vCPU, one to one-point-five GB of RAM, fifteen to twenty GB of SSD, and one to two TB of bandwidth.
Performance is acceptable for the price. Disk I/O isn't blazing—sequential reads around one hundred fifty MB/s, random IOPS around five thousand—but it's consistent. CPU steal time was moderate, around four percent.
The main advantage is cost. If you're spinning up multiple test VMs or running lightweight services, getting a year of uptime for the price of a coffee is hard to beat. Don't expect premium performance, but don't expect the server to fall over either.
BuyVM
BuyVM's Slice 512 plan is under four dollars and includes five hundred twelve MB of RAM, one CPU core, ten GB of SSD, and unmetered bandwidth at one Gbps. The disk allocation is small, but you can attach block storage volumes cheaply.
Disk performance was surprisingly good. Random IOPS consistently exceeded fifteen thousand, and sequential reads hit around five hundred MB/s. This is because BuyVM uses dedicated NVMe drives, not shared SANs.
CPU performance was clean. Steal time under one percent in all tests. The network is fast, and "unmetered" actually means unmetered—I pushed sustained traffic for hours without throttling.
The downside is availability. BuyVM sells out quickly, and you often have to wait for stock to open up. Set an alert if you want in.
What I tested and how
Each VPS was tested with the same workload. I ran sysbench for CPU, fio for disk I/O, and iperf3 for network throughput. Tests ran during peak hours (evening in the datacenter's timezone) and off-peak (early morning).
For CPU, I measured single-threaded performance and multi-threaded performance where applicable, and tracked CPU steal time using vmstat. Steal time above ten percent is a red flag. Anything under five percent is acceptable.
Disk I/O tests used fio with random read/write patterns at queue depth of thirty-two, and sequential read/write patterns at queue depth of one. I looked at both IOPS and throughput in MB/s.
Network tests measured sustained throughput to nearby targets and latency to common CDN endpoints. I also checked for packet loss over one-hour periods.
Red flags that signal overselling
High CPU steal time is the clearest indicator. Run vmstat 1 and watch the "st" column. If it's consistently above five percent, the host is overcommitted.
vmstat 1
Look at the "st" column. Numbers above ten percent mean you're waiting for physical CPU resources that are being rationed to other tenants.
Inconsistent disk I/O is another sign. Run a quick fio test:
fio --name=randread --ioengine=libaio --iodepth=32 --rw=randread --bs=4k --direct=1 --size=1G --numjobs=1 --runtime=60 --group_reporting
If IOPS varies wildly between runs or drops off a cliff during peak hours, the storage backend is saturated.
Network throttling shows up as inconsistent bandwidth. Run iperf3 to a public test server and check if throughput matches your plan's advertised uplink speed. If you're supposed to have one Gbps shared but you're capped at ten Mbps, something's wrong.
What to do after you sign up
First, run a baseline performance test immediately after provisioning. Document CPU, disk, and network numbers while the server is idle. This gives you a reference point.
Set up monitoring. Install something lightweight like netdata or use the provider's built-in metrics if available. Track CPU usage, disk I/O wait time, and network throughput over a few days.
Watch for performance degradation during peak hours. If your app is fast at three AM but crawls at eight PM, you're on an oversold node. Open a support ticket and ask to be migrated. Some hosts will move you to a less crowded server.
If performance doesn't improve, migrate to a different provider. Don't waste time fighting with a host that oversells. Your time is worth more than the two dollars you'll save by staying.
When a cheap VPS makes sense
Budget VPS hosting works well for specific use cases. Development and staging environments don't need enterprise-grade performance. You're testing code, not serving production traffic.
Personal projects and learning environments are another good fit. If you're learning Linux administration or experimenting with self-hosting, a five-dollar VPS is perfect. You get root access, full control, and if you break something, the cost of rebuilding is trivial.
Low-traffic websites and services run fine on budget VPS plans. A personal blog pulling a few hundred visits a day won't stress a single-core VPS. A private VPN endpoint for a handful of devices works great.
Backup targets and secondary services can live on cheap VPS instances. You don't need high IOPS for a backup repository that's written to once a day. You don't need low latency for a secondary DNS server.
When to spend more
Production applications with real users need better performance guarantees. If your app generates revenue or serves customers, invest in a provider with SLAs and dedicated resources.
Database servers under load will suffer on oversold hardware. Databases are disk I/O and CPU intensive. Random IOPS drop when storage is shared, and queries queue when CPU is overcommitted. Move your database to a VPS with dedicated IOPS or a managed database service.
High-traffic sites need consistent performance. If your site handles thousands of requests per hour, you can't tolerate slowdowns during peak times. You need predictable CPU and network capacity.
How do I know if my VPS is being oversold?
Check CPU steal time with vmstat or top. If the "st" value is consistently above five percent, the physical CPU is overcommitted. Run disk I/O tests with fio during different times of day. Large variations in IOPS indicate a saturated storage backend.
Can I upgrade later if performance isn't enough?
Most providers let you resize your VPS to a larger plan. The process usually involves a reboot, and your IP address typically stays the same. Check your provider's documentation for their specific upgrade process.
What's the difference between KVM and OpenVZ?
KVM is full virtualization. You get a complete virtual machine with your own kernel. OpenVZ is container-based, sharing the host's kernel. KVM gives you more control and isolation, but OpenVZ can be more resource-efficient. Most modern budget providers use KVM.
Is unmetered bandwidth really unlimited?
Usually, yes, within reasonable use policies. Providers cap the port speed (commonly one Gbps) but don't count transfer. You can sustain that speed twenty-four seven without overage charges. Read the terms of service—some hosts will suspend you for abuse, but normal hosting use is fine.
Pick based on your location and workload
Geography matters for latency-sensitive applications. Choose a datacenter close to your users. If you're serving European traffic, Hetzner or Contabo make sense. For North American users, Vultr, Linode, or BuyVM have better ping times.
Workload characteristics matter more than raw specs. A single-core VPS with fast NVMe storage will outperform a two-core VPS with slow spinning disk for database-heavy workloads. Match your provider's strengths to your application's bottlenecks.
Test before committing long-term. Most providers offer hourly billing or monthly plans. Spin up a server, run your workload for a few days, and measure actual performance. If it works, great. If not, you're out five dollars and you learned something.