8GB RAM VPS: 7 Providers Compared (And Why 8GB Is Usually Enough)

8GB is the tier most people should buy and most people overthink.

It’s the point where a VPS stops being a toy and starts being a server: enough headroom for a real database alongside a real application, enough to run a dozen containers, enough to stop watching your memory graph. It’s also the tier where the pricing gets silly — $6.99/month at Contabo, $48/month at DigitalOcean for nominally the same 8GB.

But the more useful thing to know about this tier is that at 8GB, RAM usually stops being your bottleneck. Something else takes over — and if you buy the wrong plan, that something is your disk. This guide covers both: which 8GB plan to buy, and how to tell whether 8GB is even the number you need.

If you already know you need more headroom, we’ve compared the tier above in our 16GB RAM VPS guide, where the plan shapes get genuinely strange.

The quick answer

Prices checked August 2026. Hetzner prices are in EUR and reflect its 15 June 2026 price adjustment.

The transfer column is worth reading closely: IONOS is the only genuinely unlimited entry, and a stated 20TB or 32TB allowance is often more than you’d get under an “unmetered” plan’s fair-use policy — we unpack that in our unmetered bandwidth VPS guide.

The single most important line in that table: Hostinger’s KVM 2 gives you 2 vCPUs. Every other provider here gives you 4 at the same RAM tier. More on that below, because it matters more than the price difference.

Is 8GB enough? A straight answer

Yes, for the overwhelming majority of what people self-host. Specifically, 8GB comfortably runs:

• WordPress at real traffic. Even a busy site with caching rarely exceeds 2–3GB. If your WordPress site needs more than 8GB, you have a plugin problem, not a hosting problem.
• A web app plus its database. Rails, Django, Laravel or Node alongside Postgres or MySQL, with room for Redis. This is the classic 8GB workload.
• 8–12 Docker containers doing ordinary work. Our Docker VPS guide goes deeper on sizing.
• One 7B language model, quantized. A 7B model at 4-bit needs roughly 4–5GB, which fits with room for the OS. It’ll be slow on CPU — see our Ollama hosting guide — but it fits.
• An AI agent with browser automation. Headless Chromium wants 1–2GB, so this is the first tier where agents and browsers coexist comfortably. Details in our AI agent VPS comparison.
• n8n, Coolify, Supabase, or a self-hosted app stack — each individually, with headroom.

8GB is not enough for: models above 7B, several of the above simultaneously, Elasticsearch with a large index, or anything where a single process routinely allocates multiple gigabytes. Those are 16GB conversations.

The thing nobody tells you about 8GB: disk matters most here

This is the practical difference between an 8GB box that feels fast and one that feels sluggish, and it has nothing to do with the RAM number.

Linux uses whatever memory your applications aren’t using as page cache — a read cache for your disk. On a 32GB machine running an 8GB workload, you have 24GB of cache and your disk barely gets touched. On an 8GB machine running a 6GB workload, you have well under 2GB of cache, so your database and application are hitting actual storage constantly.

At 8GB, your disk is in the hot path in a way it simply isn’t at higher tiers. Which is why the storage type in that table deserves more attention than the price:

• Contabo Cloud VPS S — 200GB NVMe. Independent benchmarking puts it around 2.1 GB/s sequential read and ~45,000 random 4K IOPS.
• Hostinger KVM 2 — 100GB NVMe.
• Hetzner CX33 — 80GB, the smallest allocation here, but consistently fast.
• Vultr Regular — 160GB SSD, not NVMe. Vultr’s High Frequency line has NVMe at a higher price.
• IONOS VPS Linux L — 240GB, the largest allocation at this tier.

If you’re running a database on an 8GB box, prioritise NVMe over an extra 40GB of capacity. The capacity is easy to add later; the latency is not.

One caveat on that advice: “NVMe” on a pricing page no longer tells you much, because providers cap IOPS and some put a network storage layer between you and the drive. Our NVMe VPS guide explains how to tell local from network storage before you buy, and gives you the fio commands to measure what you actually got.

Configure swap while you’re at it

At 8GB, a properly configured swapfile is the difference between your service degrading gracefully under a spike and the kernel’s OOM killer terminating your database at 3am. Most VPS images ship with no swap at all:

sudo fallocate -l 4G /swapfile
sudo chmod 600 /swapfile
sudo mkswap /swapfile && sudo swapon /swapfile
echo '/swapfile none swap sw 0 0' | sudo tee -a /etc/fstab
sudo sysctl vm.swappiness=10        # prefer RAM, use swap as a safety net

vm.swappiness=10 tells the kernel to avoid swapping until it genuinely needs to. Swap is not a substitute for RAM — if you’re swapping constantly, buy more memory — but as an emergency buffer it turns a hard crash into a slow minute.

The seven providers

1. Hostinger — easiest 8GB box to actually run

KVM 2: 2 vCPU, 8GB RAM, 100GB NVMe, 8TB transfer — $8.79/month on a 24-month term, renewing at $14.99.

Hostinger takes the top slot on total experience rather than raw specification. You get 100GB of NVMe, a managed control panel, free weekly backups, a browser-based terminal and one-click installs for the common stacks. For anyone who doesn’t want to spend an evening learning a cloud console, that package at $8.79 is hard to argue with.

The honest caveat, and it’s a real one: KVM 2 has 2 vCPUs where every other provider on this list gives you 4 at the 8GB tier. For a single web app with a database that’s fine — most such workloads are memory-bound, not CPU-bound. But if you’re running Docker with a dozen containers, a CI runner, or anything that parallelises, you will feel those missing cores. Hetzner’s CX33 gives you double the CPU for roughly the same money and doesn’t raise the price at renewal.

Buy Hostinger if you value the panel, the backups and the hand-holding. Buy Hetzner if you want cores.

Check Hostinger VPS pricing → · Read our Hostinger review

2. Hetzner — best specification per euro

CX33: 4 vCPU, 8GB RAM, 80GB disk, 20TB transfer — €8.49/month, flat.

Four cores and 20TB of transfer for under €9 is the best raw value at this tier, and the price is the price — no promotional rate that doubles in year two.

One caveat worth knowing before you buy: the CX line is Hetzner’s cost-optimized tier, which means it runs on older CPU generations. Hetzner says so openly — that’s how the price stays where it is — and it also means stock in this line is sometimes unavailable in your preferred region. For the workloads this tier serves, an older Xeon at 4 cores beats a newer one at 2. But if you want current-generation silicon, that’s the CPX line at a meaningfully higher price.

The 80GB disk is the smallest allocation in this comparison, which is the genuine trade-off. If you’re storing media, backups or a large database, you’ll be adding volumes (which cost extra) sooner than with Contabo’s 200GB or IONOS’s 240GB. For a typical application server, 80GB is plenty.

Hetzner also sells CAX21 — the same 4 cores and 8GB on Ampere ARM chips for €10.49. If your Docker images have ARM builds, and most mainstream ones now do, it’s worth a look. Verify your dependencies first; a stack that needs an x86-only binary will ruin your evening.

The caveats: signup sometimes triggers identity verification, the region footprint is EU-heavy with limited US presence, and the console assumes competence rather than teaching it.

The CX33 sits in Hetzner’s Cost-Optimized line, which runs on older hardware generations — that’s exactly how the price stays at €8.49. Hetzner is upfront that server availability in this line is limited.

Check Hetzner Cloud pricing → · Read our Hetzner review

3. Contabo — cheapest, with the biggest disk

Cloud VPS S: 4 vCPU, 8GB RAM, 200GB NVMe, 32TB transfer — around $6.99/month, flat.

On paper this wins outright: the lowest price, four cores, 200GB of NVMe and 32TB of transfer, with no renewal increase. Published benchmarks for this plan show genuinely quick storage — roughly 2.1 GB/s sequential reads and around 45,000 random 4K IOPS.

The consistent complaint about Contabo isn’t peak performance, it’s variance. vCores are heavily oversubscribed, so what you get depends partly on what your neighbours are doing, and two customers on the same plan can report meaningfully different results. Support response times lag the others here, and provisioning can take hours rather than seconds.

Contabo is the right answer when you want maximum resources per dollar and can absorb an occasional slow hour — a media server, a backup target, a development box, a hobby project. It’s the wrong answer for anything with a latency budget.

We don’t participate in an affiliate programme with Contabo, so there’s no tracked link here. It’s on the list because at this tier it earns its place.

4. IONOS — most storage, unlimited transfer

VPS Linux L: 4 cores, 8GB RAM, 240GB disk, unlimited transfer — $15/month, $10/month for the first 6 months.

240GB is the largest storage allocation in this comparison and the transfer is genuinely unlimited, which makes IONOS the pick if you’re serving lots of files or running something bandwidth-heavy. Four cores and real European datacentre infrastructure round it out.

At $15 standard it costs nearly double Hetzner’s CX33 for a comparable core count, so you’re paying for the storage and the bandwidth policy. The control panel is dated and slower to work in than DigitalOcean’s or Hostinger’s.

Check IONOS VPS pricing → · Read our IONOS review

5. Vultr — pay for regions, not for specs

Regular Cloud Compute: 4 vCPU, 8GB RAM, 160GB SSD, 4TB transfer — $40/month.

$40 for specs that Hetzner sells at €8.49 needs justifying, and Vultr’s justification is geography: 30-plus locations worldwide, including cities the European budget hosts simply don’t serve. If you need low latency to users in São Paulo, Johannesburg, Seoul or Sydney, Vultr’s footprint is a genuine product advantage.

Note that the Regular tier is SSD rather than NVMe — given how much disk matters at 8GB, that’s a real consideration. Vultr’s High Frequency line fixes it at a higher price.

Unless you specifically need a region the cheaper providers don’t cover, this is a lot of money for this specification. We don’t have an affiliate relationship with Vultr, so there’s no tracked link here.

6. DigitalOcean — you’re buying the platform

Basic Droplet: 4 vCPU, 8GB RAM, 160GB SSD — $48/month.

The most expensive plan here by specification, and DigitalOcean knows it. What the price buys is the surrounding platform: managed databases, Spaces object storage, load balancers, Kubernetes, a genuinely first-class API and Terraform provider, and documentation better than everyone else’s on this list.

If your Droplet is one component in infrastructure you manage as code, that ecosystem saves more engineering hours than the price gap costs. If your Droplet is your infrastructure — one box, one stack — you’re paying roughly five times Hetzner’s price for tooling you’ll never open.

Check DigitalOcean pricing → · Read our DigitalOcean review

7. Liquid Web — when you’d rather not be the sysadmin

Liquid Web’s managed VPS range starts around $33/month, and at that price you are not buying 8GB of RAM — you’re buying people. OS patching, service monitoring, incident response, and support staff who will get on a call.

Compared against $6.99 Contabo it looks indefensible. Compared against your own weekends, or against hiring someone, it’s cheap. The question isn’t which is better value per gigabyte; it’s whether server administration is work you want to be doing.

Check Liquid Web pricing → · Read our Liquid Web review

When to skip 8GB entirely

Two situations where buying 8GB is the wrong move:

Go smaller if you’re running a single service with a hosted database, a static site, a Discord bot, a headless agent calling a hosted model API, or a personal project with modest traffic. 2–4GB handles all of these, and every provider here lets you resize upward later. Our cheapest VPS comparison covers the budget end properly.

Go straight to 16GB if you plan to run a local model above 7B, several browser contexts concurrently, a full Supabase stack under real use, or an application plus database plus search index. Buying 8GB and migrating in two months costs more in your time than the price difference. The 16GB comparison has the details — and a warning, because pricing at that tier is far less rational than it is here.

How we picked

We compared published specifications and prices for plans at exactly 8GB RAM across every provider in our hosting catalogue plus the major independents, in August 2026. Where a provider sells several 8GB configurations, we listed them rather than picking the most flattering one.

We have not benchmarked these specific plans ourselves. Performance characterisations — Contabo’s storage figures and its variability, Vultr’s regional footprint, the SSD-versus-NVMe distinction — come from published third-party benchmarking and provider documentation, and we’ve noted where the picture is contested rather than settled.

Prices at this tier move often and promotional rates are aggressive. Hetzner adjusted cloud pricing effective 15 June 2026, and several comparisons still circulating online quote the older figures. Check the provider’s own page before buying.

Some links here are affiliate links, and we’ve said explicitly where a provider isn’t one. That relationship doesn’t affect the specifications or prices quoted, which come from the providers’ published pages.

FAQ

Is 8GB RAM good for a VPS?

For most self-hosted workloads, it’s the sweet spot. It comfortably runs a web application with its database, a stack of Docker containers, or an AI agent with browser automation. You’d step up to 16GB mainly for local language models above 7B, several heavy services at once, or a search index.

Why does DigitalOcean charge $48 when Contabo charges $7?

They’re selling different things under the same spec sheet. DigitalOcean’s price includes managed databases, object storage, load balancers, a mature API and strong documentation, on infrastructure with predictable performance. Contabo sells raw capacity cheaply on oversubscribed hardware with more variable performance and slower support. Both are honestly priced for what they are.

Do I need 4 vCPUs or are 2 enough at 8GB?

Depends entirely on concurrency. A single application with a database is usually memory-bound, and 2 cores are fine. Anything that parallelises — a dozen containers, a CI runner, background job workers, media processing — will use all 4 and want more. Since most providers give you 4 at this tier anyway, treat 2 cores as the thing to check for rather than the thing to pay extra to avoid.

Can I run Docker on an 8GB VPS?

Comfortably. Docker’s own overhead is minimal; what matters is what you run inside it. A dozen ordinary containers — application, database, cache, reverse proxy, a few small services — fits fine in 8GB. Set memory limits on each container so one runaway service can’t take down the whole box.

Will 8GB run a local LLM?

A 7B model at 4-bit quantization needs roughly 4–5GB and fits with headroom. Whether it’s usable depends on your patience: on CPU you’ll get a few tokens per second, which is workable for background jobs and frustrating for anything interactive. Anything larger than 7B needs 16GB or a GPU.

Is ARM worth it at this tier?

Hetzner’s CAX21 gives you the same 4 cores and 8GB on Ampere ARM for €2 more than the x86 CX33, and ARM chips are often more efficient per core. The catch is compatibility: check that every image and binary in your stack has an ARM64 build before you commit. Mainstream Docker images almost all do now; niche or commercial software often doesn’t.

How do I know if I’ve outgrown 8GB?

Run free -h and vmstat 1 5 under normal load. If the available column is consistently near zero and the si/so columns in vmstat show ongoing activity, you’re swapping and you need more RAM. If available memory is healthy but the box still feels slow, your bottleneck is CPU or disk — and more RAM won’t help.