VPS with 8 GB RAM and 4 vCPU

The Advanced plan is $13.99/mo for 4 vCPU, 8 GB RAM and 125 GB NVMe on KVM, with a 1 Gbps port, unlimited bandwidth, nightly backups kept for 3 days and one snapshot. Here is how it compares per GB of RAM with its neighbours in Devoster's range (price divided by RAM, rounded):

The pattern is plain: bigger plans are cheaper per GB, but the cheapest server is still the one whose memory you actually use. The rest of this page is about whether you will use 8 GB.

8 GB earns its price when memory is doing work that 4 GB can't hold at the same time:

• A database whose working set is a few GB. When the hot part of your data fits in the buffer pool plus page cache, reads come from memory instead of disk. That is the cheapest speed-up a database ever gets.
• Consolidating several apps or client projects onto one server, each with its own containers or PHP-FPM pool. Our guide on hosting multiple websites on one VPS covers isolating them.
• Search: Elasticsearch or OpenSearch for a small index, or Meilisearch, next to the app it serves. These lean on memory both for their own heap or index and for the OS page cache.
• Self-hosted Git with CI runners, or any server that builds Docker images, compiles code or bundles frontends while other things keep running.
• n8n in queue mode: main process, Redis and one or two workers, each worker a full Node.js process. See n8n VPS hosting.
• Collaboration suites such as a file-sync server with an online office editor, which is far heavier than the file server alone.
• Modded game servers or servers with more than a handful of players.

At Devoster, one Advanced plan is $13.99/mo and two Starter plans are $15.98/mo. Both give you 4 vCPU and 8 GB in total. The difference is in how failures and spikes behave:

Pick one 8 GB server when the workloads are related and you want the simplest thing that works, for example an app, its database and its workers. Pick two 4 GB servers when isolation matters more than $1.99 a month: production and staging, a database you don't want a noisy app next to, or two clients who should never share a failure. If the two servers talk to each other, encrypt that traffic (WireGuard or TLS), since it crosses the network.

Moving from Starter to Advanced doubles both vCPU and RAM, so it fixes either bottleneck. It is still worth knowing which one you have, because it tells you whether the next step after 8 GB should be Pro or something else entirely:

• Memory pressure: free -m shows available near zero at peak, vmstat 1 shows non-zero si/so (swapping), or dmesg shows the OOM killer. On recent kernels, /proc/pressure/memory reports how much time tasks spent waiting on memory.
• CPU saturation: the r column in vmstat 1 regularly exceeds your vCPU count while memory is fine.
• Noisy neighbours: a high st (steal) value means the host is busy, not your server. More vCPUs on the same host won't fix that. Our guide to CPU steal time explains what is normal and what to do.
• Neither: if memory and CPU are fine and the app is still slow, look at slow queries, missing indexes and external API calls before buying hardware.

8 GB goes wrong in a predictable way: every service is configured as if it owns the machine. A sample plan for a consolidated server running an app, PostgreSQL and Redis:

The 2 GB for Postgres follows the PostgreSQL documentation's suggested starting point of about 25% of RAM for shared_buffers; the OS page cache does the rest. MySQL's documentation says the InnoDB buffer pool can be set as high as 80% of memory, but that is for a server that runs nothing else. On a shared 8 GB box, 2 to 3 GB is the realistic range. Set explicit limits (maxmemory for Redis, a heap size for anything on the JVM, mem_limit in Compose) so one service can't quietly take the others' share. These figures are a starting plan, not measurements: adjust them after watching free -m through a real busy hour.

Step up to Pro (6 vCPU / 12 GB / 150 GB, $19.99/mo) or Elite (8 vCPU / 16 GB / 175 GB, $24.99/mo) when you have already tuned the memory plan above and still see pressure: several JVM services, many queue workers, headless browsers running in parallel, or a database whose working set has grown past what 8 GB can cache. Upgrades are done from the Devoster panel; the disk grows and the VPS reboots. If you are heading past the largest VPS, look at dedicated servers.

Sometimes the right move is sideways instead of up: split the database onto its own server, as in the comparison above, so each side can be sized and restarted on its own.

• You are buying RAM to hide a memory leak. A leak eats 8 GB too, just more slowly. Find it first.
• Your 4 GB server is mostly idle. If free -m shows plenty available at your busiest hour, the 4 GB VPS at $7.99/mo is the better buy.
• You need high availability. One larger server is still one server. If downtime of a single machine is unacceptable, you need redundancy, not a bigger box.
• You need a GPU, for example for AI inference. These plans are CPU-only.
• You need a managed server. Devoster VPS plans are unmanaged; the OS and software are yours to run.
• Your data must stay in Europe or Israel. The server is in Utah, USA.

To compare all eight plans, go to Devoster VPS hosting. If you're putting several sites on one 8 GB server, read how many websites a VPS can host first.