The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →For CPU-only Ollama inference, size the VPS around the model’s memory footprint, not the Ollama program itself. A practical starting point is 4 vCPUs and 16 GB RAM for a single 3B–8B model. Move to 24 GB for 13B–14B models, 24–32 GB for 27B–34B models, and at least 48 GB for a 70B model. These are estimates: quantization, context length, operating-system overhead, and concurrent users can materially change the result.
Need a server to try this on? Antihost offers Linux VPS plans with NVMe storage, full root access and a 14-day free trial.
What actually determines an Ollama VPS requirement?
Ollama’s executable uses comparatively little memory. The model you load, its quantization, and the context you give it determine most of the RAM requirement. A VPS must hold the model weights in memory while also leaving room for the operating system, Ollama’s runtime, the key-value (KV) cache used for conversation context, and any other services running on the machine.
As an Amazon Associate I earn from qualifying purchases.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →The file stored on disk is a useful minimum indicator, but it is not a complete RAM specification. A model can need more working memory than its file size, especially with a long context window or several active requests. Treat every figure below as a planning range rather than an official Ollama hardware requirement; Ollama has not published one universal RAM table that covers every model tag and quantization.
RAM requirements by model size
The following CPU-only ranges combine commonly cited VPS estimates for default 4-bit (Q4) model builds. They include practical headroom rather than claiming that the model file alone is sufficient.
#1 Best Overall
- EVOLUTION AMD RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 64GB pool, which is perfect for running LLMs such as Deepseek 32B, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 4% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
| Model parameter size | Typical RAM range | Practical VPS target | Typical use |
|---|---|---|---|
| 1B–3B | About 6–8 GB | 8 GB | Small assistants, light automation and testing |
| 7B–8B | 8–16 GB | 16 GB | General-purpose chat on one user or light traffic |
| 13B–14B | 16–24 GB | 24 GB | Higher-quality responses with moderate context |
| 27B–34B | 24–32 GB | 32 GB | Larger local models where latency is less important |
| 70B | 48–96 GB | 64 GB or more | Large-model experiments; CPU-only interaction is very slow |
A useful rule of thumb for a 4-bit model is roughly 0.6 GB of memory per billion parameters, then an additional 2–4 GB for the operating system, runtime, and context-related overhead. The rule is approximate: architecture, quantizer, context length, and the number of simultaneous sequences all affect the real result.
Model file sizes: the disk and RAM baseline
Default tags in the Ollama library provide concrete examples of how quickly storage and memory requirements grow:
- Llama 3.1 8B, 4-bit, 128K-context tag: about 4.9 GB on the library page captured on October 14, 2026.
- Llama 3.1 70B default tag: about 43 GB on that same page.
- Llama 3.1 405B default tag: about 243 GB.
- Llama 3.1 8B fp16 build: about 31 GB, illustrating how higher precision changes the requirement.
These are tag-specific values, not permanent specifications. Tags can change as models are republished, so check the current library entry before committing disk capacity. Your VPS storage must hold the downloaded model files, the operating system, Ollama data, logs, and any additional models you keep. If you retain several quantizations, add their file sizes rather than budgeting for only the model currently loaded.
Recommended VPS sizes for common Ollama workloads
One person running a 3B–7B model
Choose approximately 4 vCPUs and 16 GB RAM. An 8 GB machine can run some small 1B–3B models, but 16 GB gives an 8B-class model enough space for normal context and leaves room for the OS. NVMe storage is preferred because model loading and swapping are much less painful than with slower disks.
A single 8B model for regular interactive use
The frequently cited sweet spot is 4 CPU cores, 16 GB RAM, and NVMe storage. This is a sensible entry point for Llama 3.1 8B in a 4-bit tag. It is not a promise of a particular response speed: CPU generation remains dependent on the VPS processor, virtualization load, prompt length, and context size.
Rank #2
- 【AI MINI PC WORKSTATION】 Powered by the Intel Core Ultra 5 115U (2.70GHz base, 4.20GHz burst) with built-in Intel AI Boost NPU for local AI acceleration, this mini PC delivers efficient AI performance for daily office and creative tasks; the B11 Pro AI local computing workstation enables real-time AI tasks without cloud dependency for daily office scenarios—supporting AI photo retouching, script generation, video background blur, and real-time voice translation directly on your device for enhanced data privacy, zero latency, and offline capability.
- 【INTEGRATED GRAPHICS FOR PRODUCTIVITY】 Experience stable and smooth graphics performance with the Intel integrated Graphics GPU (boosting up to 1.8GHz), which delivers reliable office and light creative performance while maintaining low power consumption compared to entry-level desktop CPUs—this excellent power efficiency means you get desktop-class productivity performance in a silent, cool-running mini PC, with cutting-edge features like triple 8K independent display output, hardware video decoding acceleration, and full-function Type-C connectivity that ordinary compact mini PCs simply can't match.
- 【PRE-INSTALLED SYSTEM & WIDE COMPATIBILITY】 Pre-installed Windows 11 Pro OS (automatically activated online) with 13 global system languages, delivers out-of-the-box convenience for worldwide users; supports both Windows 10 and Ubuntu Linux systems, meeting the needs of office, industrial control and open-source development scenarios; compatible with mainstream office, design and conference software including Microsoft Office, Adobe Creative Suite, and Zoom, with stable performance for daily work; TPM 2.0 hardware encryption is officially supported for enterprise-level data security, alongside Windows Hello and other enterprise-grade security features.
- 【WHY LPDDR5 IS BETTER THAN DDR4】 Equipped with 16GB of onboard LPDDR5 memory running at 4400MHz, this mini PC delivers higher bandwidth and lower latency than standard DDR4 (3200MT/s). The soldered, ultra-low-power design reduces power draw and unlocks smoother multitasking, faster app loading, and significantly better integrated graphics performance—especially on Intel Core Ultra processors—so you can run multiple office software and browser tabs simultaneously and zip through daily creative workloads without stutter or slowdown.
- 【 TRANSFORM YOUR WORKSPACE WITH 8K DISPLAY SUPPORT】 Unleash unparalleled productivity by connecting three crystal-clear 8K monitors at 60Hz via HDMI 2.1, DP 2.1 and full-function Type-C port—effortlessly run stock tickers on one screen, complex spreadsheets on another, and video conferencing on the third, or dominate trading and financial modeling with real-time data sprawled across your entire field of view without any lag or stuttering.
Five concurrent users on a 7B model
One March 2026 VPS estimate places this workload at 8 vCPUs and 32 GB RAM, with an indicative cost of roughly $35–$60 per month. The price is a market estimate for that period, not an Ollama fee or a universal VPS price. Concurrent sessions consume additional KV-cache memory, and eight cores provide more scheduling capacity than a four-core personal instance.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
A 13B–14B model
Use 24 GB RAM as the comfortable target. A 16 GB VPS may start the model with a short prompt, yet leave too little room for a long context, background services, or more than one request. Keep at least four vCPUs for a single interactive user and increase CPU capacity if several users will generate simultaneously.
A 27B–34B model
Plan on 32 GB RAM. The lower end of third-party estimates is around 24 GB, but 32 GB avoids running at the edge of capacity and gives the KV cache and operating system room to breathe. CPU-only generation will be slower than with an 8B model.
A 70B model
Estimates range from 48 to 96 GB RAM. That wide spread reflects different assumptions about quantization, context, and headroom. A 64 GB VPS is a more realistic starting point than 48 GB if you want a usable margin, while 96 GB is safer for long contexts or multiple users. The 43 GB default 70B file does not mean a 43 GB VPS is sufficient.
CPU requirements and expected speed
On a CPU-only VPS, throughput is usually the limiting factor after the model fits in memory. A 2026 third-party comparison estimated roughly 8–20 tokens per second for 7B models and about 0.8–3 tokens per second for 70B models. These ranges vary with the host CPU, thread allocation, quantization, prompt length, and system load; they are not guaranteed benchmarks for every provider.
Recommended Free Tools
Rank #3
- Powerful AI Processor: Experience next-generation AI technology, greatly improve productivity, and bring unprecedented high peraformance with the latest AMD Ryzen Al 9 HX 370 processor (Up to 5.1 GHz, 12 Cores / 24 Threads). With the support of AMD Radeon 890M, you can play your favorite AAA games with smooth, stunning graphics and zero latency.
- Intelligent AI Assistant: Mini PC AI X1 Pro has a built-in new Copilot AI function and supports Recall function - just describe the details in your memory to retrieve the content you have recently browsed or used. At the same time, the built-in real-time subtitle translation provides subtitles simultaneously during video calls or watching movies. Press the dedicated Copilot button to activate the AI assistant in Windows 11, quickly answer questions, inspire creativity and improve work efficiency. In addition, the fingerprint sensor realizes fast and secure unlocking.
- Extreme audio experience and efficient noise reduction: Equipped with dual noise reduction DMIC and built-in speakers, you can enjoy clear and noise-free sound quality experience in video conferencing, audio and video entertainment and voice interaction. The audio system and AI assistant work seamlessly together to ensure intelligent and efficient workflows.
- High-speed connection and strong expansion performance: Equipped with dual USB4 interfaces to ensure fast and unimpeded data transmission and support connecting to eGPU through the OCuLink port, opening up a super-smooth gaming experience and a stunning visual feast. Supports three ultra-fast PCIe 4.0 SSDs(Total 1TB), supports a loading speed of up to 7000MB/s, and can be expanded to up to 12TB of storage; it is also equipped with up to 32GB 5600MHz DDR5 removable memory (up to 128GB), allowing multitasking with ease.
- Intelligent Cooling Design & Energy Saving: The CPU and SSD are equipped with independent fans, while the memory and built-in power supply feature an efficient heat dissipation design. This setup ensures enhanced thermal management throughout the system. Even under high load conditions, it maintains a full-load noise level as low as 45dB and keeps maximum power consumption at 65W. Additionally, the built-in 135W power adapter minimizes stability issues and noise associated with external power adapter connections.
For a personal 7B–8B deployment, four vCPUs is a reasonable baseline. More cores can improve throughput or allow several requests to progress at once, but adding vCPUs cannot make a memory-starved instance work. If the model is swapping, reduce the model or increase RAM before paying for more CPU.
A 70B model at under a few tokens per second can take a long time to produce an ordinary answer, making it impractical for fast, game-like dialogue or an interactive coding companion on CPU alone. A GPU can substantially change latency, but GPU requirements depend on the model’s quantization and the amount of weights that fit in VRAM; the figures in this guide cover CPU-only VPS planning.
Why quantization changes the answer
Quantization stores weights at lower numerical precision. A 4-bit tag is much smaller than an fp16 tag, which is why an 8B model can occupy about 4.9 GB in one 4-bit library tag but about 31 GB in fp16. Higher precision can improve fidelity for some tasks, yet it raises disk use and working-memory demand sharply.
When comparing VPS plans, identify the exact tag, not only the model family and parameter count. “Llama 3.1 8B” does not uniquely describe the memory footprint if 4-bit, 8-bit, and fp16 variants are available. Downloading a larger-precision tag than planned can turn a comfortable 16 GB instance into an out-of-memory system.
Context length and KV-cache memory
Context length controls how much conversation, retrieved text, or source code the model can consider at once. Longer contexts require a larger KV cache, so two users running the same model can consume very different amounts of RAM depending on prompt history and the configured limit.
Rank #4
- 【AI-Accelerated Processor】AI X1-470 mini pc equipped with an AMD Ryzen AI 9 HX 470 processor (up to 5.2 GHz, 12 cores, 24 threads), this system delivers local AI performance of up to 86 TOPS. This enables low-latency AI workloads directly on the device, reducing reliance on the cloud and providing reliable computing power for productivity and intelligent applications.
- 【Workstation-Level Graphics Expansion】Integrated Radeon 890M graphics supports demanding creative tasks and modern games, while OCuLink (via M.2 adapter) enables external desktop GPU expansion for high-end rendering and advanced visual workloads, providing scalable graphics performance as needs grow.
- 【Quad 4K Display & High-Speed Connectivity】Mini computer X1-470 equipped with USB4(High-speed data transmission, video output, and power supply can be achieved through a single cable.), HDMI 2.1 FRL, DP 2.0, Wi-Fi 7, and 2.5GbE LAN, this mini PC supports up to four 4K displays and high-bandwidth peripherals, ideal for multi-screen trading, creative production, and professional office setups without requiring external docking stations.
- 【Massive DDR5 Memory & Dual M.2 Storage】Supports up to 128GB DDR5 memory and dual M.2 SSD expansion up to 8TB, ensuring smooth multitasking, large AI model execution, and high-resolution video editing without storage or memory bottlenecks.
- 【Advanced Cooling & Integrated Audio System】Featuring phase change material, dual copper heat pipes, and active cooling design, the system maintains stable performance under heavy workloads (full-load temperature under 80°C, noise under 45dB), while built-in noise-reduction microphones and speakers enhance video conferencing and AI voice interaction efficiency.
The 4.9 GB Llama 3.1 8B example is identified with a 128K context tag. That label describes the model’s supported context window; it does not mean a 16 GB VPS can hold 128K tokens for many simultaneous sessions without additional memory. If you do not need a huge history, setting a smaller context limit is one of the simplest ways to reduce memory pressure.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to size an Ollama VPS step by step
- Choose the exact model tag. Record the parameter count, quantization, and advertised context length. Use the current Ollama library entry because tag sizes can change.
- Start with the model’s file size. Use it as a lower bound for storage and a warning that RAM must be larger than the file. A 43 GB 70B file, for example, rules out a 32 GB machine before the service is installed.
- Add operating-system and runtime headroom. Reserve approximately 2–4 GB beyond a 4-bit model as a practical minimum margin, then increase that margin for long contexts or background services.
- Account for concurrency. Each active conversation has its own prompt and KV-cache needs. Size RAM for the maximum simultaneous generations, not the average number of logged-in users.
- Choose CPU cores for latency goals. Use about four vCPUs for one light interactive workload; move toward eight vCPUs when several users share a 7B model. More cores do not replace missing RAM.
- Provision sufficient disk. Include the OS, Ollama data, model files, logs, and every additional tag you intend to keep. NVMe storage makes initial downloads and model loading less frustrating.
- Test the real prompt pattern. Measure response speed and memory while using the context length and concurrency you expect. A short single-user test can hide the cost of long histories or parallel requests.
Examples: choosing between common deployments
Personal game-development assistant
For one developer asking an 8B model about scripts, quest logic, or engine errors, select 4 vCPUs and 16 GB RAM with NVMe storage. Keep the context limit realistic for the tasks; pasting an entire project into every prompt creates a larger KV cache than short question-and-answer use.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchSmall team chatbot
For up to five people sharing a 7B model, use the 8-vCPU, 32-GB pattern cited for concurrent use. Monitor memory while several users submit long prompts at once. If requests queue rather than complete concurrently, CPU capacity or a request limit may be the constraint; if the kernel begins reclaiming or swapping memory, RAM is the constraint.
Quality-focused 14B deployment
A 24 GB VPS is the practical target for a 13B–14B 4-bit model. Keep other services lightweight and avoid loading a second large model at the same time. If you need long contexts and several users, move above the baseline instead of relying on swap.
Large-model experiment
A 70B model requires at least 48 GB by the broadest estimates and is more comfortable with 64–96 GB. Expect CPU-only output to be slow, commonly around 0.8–3 tokens per second in third-party measurements. This setup is for experimentation or batch jobs unless your users can tolerate long waits.
Common sizing mistakes and fixes
- Matching RAM to the download size: a model file is not the complete runtime footprint. Increase RAM and leave room for context and the OS.
- Ignoring quantization: switching from 4-bit to fp16 can multiply storage and memory needs. Verify the tag before downloading.
- Using a maximum context by default: long contexts increase KV-cache use. Lower the context limit when your application does not require it.
- Planning for average rather than peak concurrency: five users arriving together need more memory and CPU than one user tested alone.
- Trying to cure out-of-memory errors with more vCPUs: cores improve scheduling and throughput, but they do not create RAM. Upgrade memory or choose a smaller quantized model.
- Relying on swap for interactive inference: swap can prevent an immediate crash but usually makes generation unusably slow. Treat it as an emergency safety net, not capacity.
- Assuming a 70B model will feel like a 7B model: parameter count affects both memory and CPU throughput. The larger model may fit yet still be unsuitable for interactive response times.
Quick decision guide
| Goal | Suggested starting point | Important qualification |
|---|---|---|
| Small 1B–3B model | 4 vCPUs, 8 GB RAM | Use a larger plan for long context or background services |
| Single 7B–8B model | 4 vCPUs, 16 GB RAM, NVMe | 16 GB is a common target, not an official minimum |
| Five concurrent 7B users | 8 vCPUs, 32 GB RAM | March 2026 estimate; roughly $35–$60/month at that time |
| 13B–14B model | 4 or more vCPUs, 24 GB RAM | Increase memory for long context or multiple sessions |
| 27B–34B model | 32 GB RAM | CPU-only latency is higher than with 7B–8B models |
| 70B model | 64 GB RAM as a practical starting point | Published estimates span 48–96 GB; CPU-only speed is commonly 0.8–3 tokens/s |
Bottom line for 2026 VPS planning
An Ollama VPS should have more RAM than the model file occupies, with extra capacity for context and the operating system. For most people, an 8B 4-bit model on a 4-vCPU, 16-GB NVMe VPS is the sensible entry point. Choose 24 GB for 13B–14B models, about 32 GB for 27B–34B models, and 48 GB or more for 70B models—preferably 64 GB or higher if you want useful headroom. Confirm the exact tag, quantization, context length, and peak number of users before ordering, because those variables determine whether the VPS is merely able to load Ollama or can serve it responsively.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
