Skip to main content
Servers Are Up
Virtual Datacenter
PricingFree Tools
Sign inBrowse catalog
Servers Are Up
Virtual Datacenter
PricingFree Tools
Sign inBrowse catalog
Servers Are Up

Your apps, automation and data on hardware we own. Build a private stack with predictable capacity and access through your dashboard, API or SSH.

SLA with service creditsOwned hardwareUS-based

Make room for your next idea

Explore the capacity, apps and access options that fit your workload. Check current availability and your total before you buy.

Sign up freeView Pricing

Products

  • Linux VPS
  • Windows VPS
  • Cloud & SaaS Services
  • Cloud Storage (Azure)
  • Virtual Datacenter
  • AI & Agent Hosting
  • Containers
  • Instant Apps
  • Software Platforms
  • Network builder
  • Game Servers
  • Pricing

Resources

  • Free Network Tools
  • 2026 Pricing Benchmark
  • Live SVG Badges
  • Latency & Speed Test
  • Guides & Tutorials
  • Documentation
  • API Reference
  • MCP Server
  • CLI & SDKs
  • OpenAPI & llms.txt
  • Use Cases
  • Free Migration
  • Uptime & Reliability
  • Datacenter
  • System Status
  • Network
  • Provider Comparisons Hub
  • Compare Providers

Company

  • About Us
  • Enterprise
  • Security Architecture
  • Trust Center
  • Blog
  • FAQ
  • Compliance
  • Startup Program
  • Open-Source Sponsorship
  • Press & Media Kit
  • Referral Program
  • Help Center
  • Contact Us
© Servers Are Up
TermsPrivacySLARefund PolicyDPAAcceptable Use
Service information
Back to Engineering BlogSystems, Linux & KVM
Servers Are Up Engineering·Published 2026-11-22·12 min read·Technical Review Verified

Server Sizing for Workloads: How to Pick CPU, RAM and Storage

server sizing for workloads choose server size how much ram does a server need

Engineering Summary & Direct Answer

Size a server to the workload, not the price tier: measure peak load, set RAM, CPU, storage and network separately, add headroom and know when to resize.

Instant 60s deploy · Browser SSH & WebRDP included
Deploy Linux VPS Compare Full PricingVirtual Datacenter Pools ($12–$380/mo)
On this page (13 sections)
  1. TL;DR
  2. Why size to the workload instead of the plan name
  3. Step 1: Describe the workload before you look at specs
  4. Step 2: Measure peak usage, not the average
  5. Step 3: Size each resource on its own
  6. RAM
  7. CPU
  8. Storage capacity and storage speed
  9. Network
  10. Step 4: Add headroom, and know what kind
  11. Sizing by workload
  12. Web apps, APIs and business services
  13. Game servers
  14. AI projects
  15. Windows desktops and servers
  16. When to resize
  17. Our view: size small, measure, then grow
  18. Put the sizing into practice
  19. Frequently Asked Questions
  20. How much RAM does a server need?
  21. How do I choose a server size if I have no data yet?
  22. Is it better to get a bigger server or several small ones?
  23. How do I know my server is oversized?
  24. Sources
  25. Recommended

The right server size is the smallest configuration that still handles your busiest hour with room to spare. You find it by measuring one workload at its peak, sizing CPU, RAM, storage and network separately, and adding headroom. Picking a plan by name or by price alone is how people end up either paying for idle cores or watching a process get killed for lack of memory. On Servers Are Up you can pick a size per workload across Linux, Windows, containers and game servers, and manage every machine from one browser tab.

This guide walks through the method, then applies it to the four workloads most readers here run: web apps and APIs, game servers, AI projects and Windows desktops.

TL;DR

  • Size for peak load, not average load. A server that idles at 10% CPU can still fall over at the daily spike.
  • CPU, RAM, storage and network each fail on their own. Check all four, and storage speed as well as storage capacity.
  • RAM is the resource that fails hardest: when it runs out, processes are killed or the machine slows to a crawl. Size it first and leave a margin.
  • Start from a reasonable tier, watch real usage for a couple of weeks, then adjust. Sizing is a loop, not a one-off guess.
  • Every workload below maps to a Servers Are Up plan you can size per unit, from Linux Starter at $11 per month up to pooled resources for teams.

Why size to the workload instead of the plan name

Plan names (Starter, Standard, Pro) are labels for bundles of CPU, memory and disk. They tell you nothing about whether the bundle matches what you run. Two apps that both "serve a website" can differ tenfold in memory use: a static site behind a cache barely touches RAM, while a Java service with a large heap or a database with a big working set can take most of it on its own.

Undersizing and oversizing both cost money. Too small, and you get slow responses, restarts and outages at the worst moment. Too large, and you pay every month for capacity nobody uses. The goal is a server whose busiest hour lands somewhere in the comfortable middle of its limits. For a deeper look at the forecasting side, our server capacity planning guide covers growth over months rather than a single sizing decision.

Step 1: Describe the workload before you look at specs

Write down four facts before comparing any sizes:

  1. What runs on it. One web app, a database, a game world, a model server, a Windows desktop for several people?
  2. How many people or requests hit it at once. Concurrency matters more than total monthly traffic. Ten thousand visitors spread over a day are easy. Ten thousand in ten minutes are not.
  3. When the peak happens. A game server peaks on weekend evenings. A business app peaks at 9 a.m. A nightly backup or build job can create a peak nobody planned for.
  4. What you can tolerate. A hobby project can be slow for a minute. A customer checkout cannot.

This short brief decides everything that follows. Without it, any number you pick is a guess.

Step 2: Measure peak usage, not the average

Averages hide the problem. A server averaging 25% CPU over a day may sit at 95% for twenty minutes every afternoon, and those twenty minutes are what your users remember. Capacity guidance consistently recommends recording the metrics that show the limit: response-time percentiles such as P95 and P99, error rate, throughput, and CPU, memory, disk and network use at each load step. Percentiles matter because they show the slowest requests, which averages smooth away.

If you already run the workload somewhere, collect at least a couple of weeks of data that includes your busiest days. If it is new, run a load test that ramps traffic step by step and note where things bend. The same guidance points out that a resource hitting full utilization, errors staying above your threshold, growing P99 latency or throughput that stops rising all mark the capacity limit. Correlate them: if latency climbs exactly when CPU flattens at 100%, CPU is your bottleneck.

Two practical habits help:

  • Run a long test, not only a short burst. Memory that keeps climbing without leveling off can be a leak, and a sustained run can expose problems a short test misses.
  • Test with realistic data. An empty database fits in memory. A year of production data may not.

Step 3: Size each resource on its own

A server is four constraints, not one. Any of them can be the limit while the others sit idle, so check each separately.

RAM

RAM is the resource that punishes mistakes most. CPU shortage makes things slower. Memory shortage makes things stop: the operating system starts swapping to disk, which is far slower than memory, or kills the biggest process outright.

To estimate RAM, add up what each piece needs at peak:

  • The operating system and background services.
  • Each application process or runtime, including its heap or cache settings.
  • Your database's working set, meaning the portion of data it wants in memory to answer queries quickly.
  • File system cache, which the operating system fills with spare memory and which speeds up reads.
  • Headroom, so a traffic spike does not push you over the edge.

The signs that you are short on memory are consistent: swap use rising, out-of-memory kills in the system log, and latency that grows with uptime. If you see memory growth that never plateaus, look for a leak before you buy more RAM.

A useful starting method is to run the real workload for a few days, read peak resident memory for every process, add them together, then add a margin on top. Java services deserve special care because the runtime reserves a heap plus memory outside it; our JVM memory guide breaks that down.

CPU

CPU needs are about how much work happens at the same time. A single-threaded program cannot use eight cores, so more cores do not help it, while a web server handling many parallel requests scales with core count. Ask two questions: is the workload parallel, and does it have one hot thread? Game servers are the classic example, where one main thread often sets the ceiling and clock speed matters as much as core count.

Watch sustained CPU use at peak. If it sits near full for long stretches, and response times rise at the same moment, you need more compute or a lighter workload. If peak CPU is under half, a smaller size will probably do.

Storage capacity and storage speed

Capacity is simple: operating system, application, data, logs, backups and growth. Logs and backups are the usual surprise, because they grow quietly until the disk fills. Add storage as a separate line item when you need it. On our platform, storage is a paid add-on, for example the 512 GiB standard storage drive, so you can keep the server itself small.

Speed is the part people forget. A database doing many small reads and writes can be limited by disk operations per second long before the disk is full. Look at disk I/O wait during your peak, not just free space.

Network

Network limits show up as bandwidth caps, connection counts and latency. Streaming, file distribution and big game lobbies use bandwidth. APIs with many small calls care more about latency and connection handling. If the machine needs a fixed address, a public IPv4 address is $5 per month on top of the server. Resilience belongs here too: the way you test network redundancy and failover decides whether one link problem becomes an outage. Our guide to virtual network management covers planning and securing the network side.

Step 4: Add headroom, and know what kind

Headroom is the margin between your measured peak and the server's limit. It absorbs growth, unexpected spikes and the overhead you did not measure. Sizing guidance describes the trade-off plainly: too little margin risks availability, too much wastes budget.

A workable approach:

  • Leave visible spare capacity on RAM, since running out is the harshest failure.
  • Keep CPU peaks well below saturation so latency stays flat.
  • Plan for the growth you can see in the next few months, not the growth you hope for years out.
  • Do not rely on memory overcommitment for production. Giving several virtual machines more memory in total than the host has works until they all want it at once.

If you split one big machine into several virtual ones, total every memory allocation and add the host's own overhead before deciding the pool is big enough.

Sizing by workload

The table maps common workloads to the resource to watch first and the Servers Are Up plan family where they usually live. Treat the starting points as where you begin measuring, not as final answers.

Workload Watch first Starting point
Small web app or API RAM, then CPU Linux Starter, $11 per month
Production web app with a database RAM and disk speed Linux Standard, $38 per month
Heavier services, several apps on one machine RAM and CPU together Linux Pro, $76 per month
Containerized apps Per-container memory limits Docker Starter, $11 per month
Kubernetes development Node memory, scheduling K8s Dev Cluster, $38 per month
Windows workloads RAM, CPU Windows Starter, $35 per month
Game server Main-thread CPU, RAM Game servers from $7.99 per month
AI agents RAM, API calls Agent Micro, $9 per month

Web apps, APIs and business services

Start with RAM for the runtime and database, then confirm CPU under a load test. Most web workloads are limited by memory and database speed before raw compute. If you run several small services, containers let you give each one its own limit instead of buying a separate machine for every one. Our Docker Compose linter helps you check a compose file before you deploy it.

Game servers

Game servers are sized by the game, the player count and any mods. A modded world with many players needs more memory and stronger single-thread performance than a small vanilla one. We publish a RAM, CPU and tickrate sizing guide for 34 multiplayer games, which is the best place to look up your title. Then pick from our game server hosting options.

AI projects

AI sizing starts with the model. When a model runs on a GPU, the guidance from VMware's AI sizing document says GPU-enabled virtual machines need memory for the operating system and loaded software, and that a properly functioning modern large language model should not swap into VM RAM. It also offers a rule of thumb of two to three times the total GPU memory as server RAM, while noting that overhead and other workloads need their own allowance. For hobbyists running agents that call a model over an API rather than hosting one, the server mostly needs steady memory and a reliable connection. Our CPU vs. GPU explainer helps you decide which one your project needs.

Windows desktops and servers

Windows carries a larger baseline than a minimal Linux install, so it needs more memory to do the same job. That is why the Windows tiers start higher: Windows Starter is $35 per month, Standard $66 and Pro $116. If several people will work on one desktop at once, size for the total of their sessions, not one user. The Windows Server desktop is $66 per month.

When to resize

Resize on evidence, not on a feeling. Signs the server is too small:

  • Memory use near the limit at peak, with swap growing or processes being killed.
  • CPU pinned at full utilization while response times climb.
  • Disk I/O wait rising during busy periods.
  • Errors that stay above your threshold under normal load.

Signs it is too big: CPU and memory both stay low even on your busiest day, for weeks. Then step down a tier and keep the savings.

Revisit sizing after any change that alters the load: a new feature, a marketing push, a bigger player base, a larger dataset or a model upgrade. Server uptime also depends on routines around the machine, which we cover in our uptime best practices.

Our view: size small, measure, then grow

We build Servers Are Up around the idea that sizing should be a routine you can repeat, not a bet. Every server, container and network lives in one browser console, so you can open a terminal and a desktop side by side and watch memory and CPU on a machine while you change the thing that is using them.

Because sizing is per unit, you can begin with the tier that fits today's measured peak and move when the numbers say so. Our servers run on hardware we own and operate in the United States, which means the capacity you size against is capacity we manage directly.

For teams, pooled resources fit the same idea: a shared allocation you divide among machines, with an API and an MCP endpoint so your own AI agents can read usage and automate the next step.

— Servers Are Up Engineering

Put the sizing into practice

Once you have a measured peak and a margin, the choice is quick. Pick a Linux or Windows server for a single workload, containers for several small services, or a virtual datacenter when you want a pool of resources to divide up: Hobby Lab is $32 per month, Developer Studio $64 and Team Workspace $258. Machines you already run elsewhere can be connected for $3 per month per machine, so one console can show everything.

See every plan and size on Servers Are Up and choose the one that matches your numbers.

Frequently Asked Questions

How much RAM does a server need?

It depends on what runs on it, so add up peak memory for the operating system, your application, the database working set and cache, then add a margin. Measure over a couple of weeks that include your busiest period. RAM is the resource to be most generous with, because running out stops processes rather than just slowing them.

How do I choose a server size if I have no data yet?

Start with the smallest tier that fits the software's stated requirements, then run a load test that ramps up step by step. Watch where latency bends or a resource fills. Move up one tier at a time and repeat the test until the peak sits comfortably below the limit.

Is it better to get a bigger server or several small ones?

It depends on the workload. A single-threaded or tightly coupled program needs one strong machine. Stateless services spread easily across smaller machines or containers, which also lets you scale in small steps. Containers and pooled resources on our platform suit that second case.

How do I know my server is oversized?

If CPU and memory stay low on your busiest days for several weeks, and latency is flat, you are paying for unused capacity. Drop one tier, keep watching the same metrics and confirm the peak still has room.

Sources

  • Capacity Testing Guide: Load Limits & Scalability, QAPractices
  • Sizing AI Workloads on VCF, VMware
  • Network Redundancy: Test Failover for Businesses, Vuetek Systems

Recommended

  • Server Capacity Planning: A Practical Guide to Sizing and Forecasting
  • Best Dedicated Game Server Hosting in 2026: Complete RAM, CPU & Tickrate Sizing Guide for 34 Multiplayer Games
  • 2026 Cloud VPS Price-Per-GB RAM Benchmark: DigitalOcean vs Vultr vs Linode vs Owned US Hardware
  • Project Zomboid Dedicated Server Hosting Guide: RAM, JVM Tuning, Mods & Ports (Build 41/42)

US-Owned Dell PowerEdge R940 KVM Infrastructure

Test & Run Production Workloads on Predictable KVM Compute

Linux VMs from $6/mo · 4 vCPU / 8 GB ECC RAM $18/mo · 8 vCPU / 16 GB RAM $28/mo · Pooled Virtual Datacenters from $12/mo with built-in Browser SSH & WebRDP.

Deploy Linux VPS Compare Full Pricing

Explore With AI & Cite This Page

Open a grounded prompt in your AI assistant or copy a clean Markdown, HTML, or APA reference link.

Ask ChatGPT Ask Perplexity Ask Claude
Share:HN·Reddit·X·LinkedIn
Older Article

What Is Virtualization Technology? A Practical Guide

Related Diagnostic Tools & Engineering Guides

Validate protocols and subnets directly in your browser or continue reading related architecture notes.

Free Browser Tool

Free WHOIS & RDAP Domain & ASN Lookup Tool

Launch utility
Free Browser Tool

SSH Keygen Generator & Cloud-Init YAML Builder

Launch utility
Free Browser Tool

CVE Vulnerability & CISA KEV Banner Advisor

Launch utility

Server Capacity Planning: Little's Law, Queuing Knee Math, Linux PSI Telemetry, and Tier Sizing

Size servers quantitatively using Little's Law, M/M/1 queuing utilization knees, Linux Pressure Stall Information (/proc/pressure), and working-set RAM math.

Game Server Control Panel: What It Does and How to Choose

What a game server control panel does, which features matter, how to set one up and how to choose between panel types, with a browser-based option from $7.99 a month.

Browser-Based SSH Terminal Setup for Multi-Server Access: WebSocket Bastion & Guacamole Architecture

Architectural deep dive into multi-server browser SSH and RDP workspaces: xterm.js WebGL rendering, single-use WebSocket session tickets, loopback guacd multiplexing, and SSH key hygiene.

How to Install Python on Windows 11 & Windows Server 2025: uv, WinGet, MAX_PATH & NSSM Services

Modern 2026 guide to installing and running Python 3.13+ on Windows 11 and Windows Server 2025: Astral uv vs winget + py.exe launcher, PowerShell execution policy, fixing MAX_PATH >260 chars, and deploying 24/7 services via NSSM.

On this page

  1. TL;DR
  2. Why size to the workload instead of the plan name
  3. Step 1: Describe the workload before you look at specs
  4. Step 2: Measure peak usage, not the average
  5. Step 3: Size each resource on its own
  6. RAM
  7. CPU
  8. Storage capacity and storage speed
  9. Network
  10. Step 4: Add headroom, and know what kind
  11. Sizing by workload
  12. Web apps, APIs and business services
  13. Game servers
  14. AI projects
  15. Windows desktops and servers
  16. When to resize
  17. Our view: size small, measure, then grow
  18. Put the sizing into practice
  19. Frequently Asked Questions
  20. How much RAM does a server need?
  21. How do I choose a server size if I have no data yet?
  22. Is it better to get a bigger server or several small ones?
  23. How do I know my server is oversized?
  24. Sources
  25. Recommended
Self-Serve Infrastructure

Test & Run Production Workloads on Predictable KVM Compute

Linux VMs from $6/mo · 4 vCPU / 8 GB ECC RAM $18/mo · 8 vCPU / 16 GB RAM $28/mo · Pooled Virtual Datacenters from $12/mo with built-in Browser SSH & WebRDP.

Deploy Linux VPS Compare Full Pricing