For a small Ollama model on a CPU-only VPS, start your shortlist with Hostinger KVM 2 at $8.79 per month equivalent, billed as $210.96 for 24 months. Choose KVM 4 at $12.99 per month equivalent if you want 16 GB RAM for more room to load a quantized 7B model and the operating system. Neither plan includes a listed GPU, so expect CPU inference and validate speed before moving a real service.
Affiliate disclosure: Some links are affiliate links. We may earn a commission on qualifying purchases at no extra cost to you. Read our affiliate disclosure .
Based on official provider pricing and Ollama documentation checked November 18, 2026. Model fit guidance is an editorial sizing estimate, not a performance test.
Compare the VPS options for Ollama
An Ollama server needs room for Linux and its services, the model files on disk, and the model plus runtime data in memory. More RAM does not make a CPU-only server behave like a GPU machine. Use the table to decide whether a low-cost experiment is enough or your use case needs dedicated accelerator hardware.
| Plan | Intro price and first term | Renewal shown | CPU / RAM / storage | Ollama fit |
|---|---|---|---|---|
| Hostinger KVM 1 | $6.49/month equivalent; $155.76 for 24 months | $11.99/month for 24 months | 1 vCPU / 4 GB / 50 GB NVMe | A 1B or small 3B quantized model for low traffic and short context; little room for other services |
| Hostinger KVM 2 | $8.79/month equivalent; $210.96 for 24 months | $14.99/month for 24 months | 2 vCPU / 8 GB / 100 GB NVMe | Best starting balance for small CPU-only models and a simple API or reverse proxy |
| Hostinger KVM 4 | $12.99/month equivalent; $311.76 for 24 months | $28.99/month for 24 months | 4 vCPU / 16 GB / 200 GB NVMe | More memory for a 7B quantized model at modest context, but CPU generation can still be slow |
| UltaHost VPS Professional | $15.87/month; $380.88 calculated for 24 months | $15.87/month for 24 months | 3 CPU cores / 4 GB / 75 GB NVMe | A managed-server alternative for a smaller model; 4 GB limits model and operating-system headroom |
Hostinger bills the listed introductory prices upfront for 24 months. UltaHost displays the rate for its two-year term; the $380.88 total is calculated as $15.87 x 24. Rates exclude checkout changes and taxes. Confirm the selected term and total before paying. Sources: Hostinger official VPS plan pricing , checked November 18, 2026; UltaHost official VPS plans , checked November 18, 2026.
What the memory and model sizes mean
Model download size is only a starting point for RAM planning. Ollama’s library lists the default Llama 3.2 3B quantization at 2.0 GB, while the Qwen2.5 7B instruct Q4_K_M file is 4.7 GB. The model also needs runtime memory, and the operating system, Ollama, context length and any proxy or database share the server’s RAM. Leave extra space instead of treating the model file size as the full requirement.
Ollama documents that context length can be configured and that concurrent requests consume more memory. Its FAQ says parallel request memory scales with OLLAMA_NUM_PARALLEL multiplied by OLLAMA_CONTEXT_LENGTH. Start with one request at a modest context, then inspect actual memory use before increasing either setting. A model card advertising a very long context does not mean a small VPS can run that context efficiently.
Sources: Llama 3.2 model sizes in Ollama’s library , Qwen2.5 7B instruct model details , and Ollama FAQ on context and concurrency , checked November 18, 2026.
Why Hostinger KVM 2 is the practical starter
KVM 2 has 8 GB RAM, two vCPU cores and 100 GB NVMe storage in the current plan listing. That gives a small CPU-only Ollama setup more operating room than a 4 GB entry tier while keeping the initial 24-month price below the 16 GB option. It is a reasonable shortlist for trying a compact quantized model, a private API and light personal use.
Choose KVM 1 only if you are deliberately keeping the setup small: one compact model, low concurrency and no busy database or unrelated containers. KVM 4 is the more comfortable Hostinger choice for loading a 7B quantized model because it has 16 GB RAM, but four vCPU cores are still CPU resources. A larger memory tier can prevent memory pressure; it does not promise acceptable tokens per second.
For a plan-by-plan RAM discussion, read how much RAM a VPS needs . If your application is an API, database and several containers around the model, our Docker VPS comparison helps estimate the rest of the stack.
Start with Hostinger VPS plans for your Ollama workload
When a CPU-only VPS is the wrong purchase
Pick a GPU server or a hosted inference API if response time, repeated generation, larger models or multiple users matter. Ollama’s GPU guidance explains that available VRAM is used when scheduling models across supported GPUs. The Hostinger KVM plan cards compared above list CPU, RAM and NVMe storage, not an attached GPU. Do not buy one expecting GPU acceleration.
CPU-only hosting can still make sense for learning, private low-volume automation, scheduled jobs or a small model behind an internal tool. Keep the first deployment narrow. Test the exact model, quantization, prompt length and request pattern you plan to serve. If generation is too slow, compare the cost of a GPU instance with an API bill rather than moving blindly to a larger CPU tier.
An always-on AI agent VPS may only need to host the agent process while calling an external model API. That is a different, lighter workload than keeping Ollama’s model in memory on the same server.
Who should choose which option?
- You are learning Ollama with a compact model: start by evaluating Hostinger KVM 2. It has 8 GB RAM for the model and supporting services, but test latency before exposing it to users.
- You need the lowest introductory bill: consider KVM 1 only for a small model, short context and low concurrency. Its 4 GB RAM leaves less room for the OS and supporting services.
- You want to try a 7B quantized model on CPU: shortlist KVM 4 for 16 GB RAM and 4 vCPU, then measure whether generation speed is useful for your task.
- You need interactive speed, larger models or several users: price a GPU instance or API first. The CPU VPS plans here do not list a GPU.
- You value a managed-server offer more than RAM per dollar: compare UltaHost Professional, but its 4 GB allocation makes it a small-model option rather than a clear fit for larger local models.
Before you pay
Write down the model and quantization, intended context length, number of simultaneous users, storage for downloaded models and the services that will share the host. Check the provider’s current term and renewal bill. Then install Ollama, run your representative prompt privately, and watch system memory and response time under the actual workload. Do not expose the Ollama API publicly without an access-control and network plan.
FAQ
Can Ollama run on a VPS without a GPU?
Yes. Ollama can use system memory for CPU inference. A CPU-only VPS is most appropriate for learning or low-volume use with a compact model. It does not provide the VRAM acceleration of a supported GPU.
Is 8 GB RAM enough for Ollama?
It can be a sensible starting point for smaller quantized models, but model file size is not total memory use. Context length, concurrency, Linux and other services all need memory. Check the specific model and monitor the server before increasing workload.
Can I run a 7B model on Hostinger KVM 2?
The Qwen2.5 7B Q4_K_M model file is listed at 4.7 GB, while KVM 2 has 8 GB RAM. That leaves limited headroom for runtime, context and Linux, so KVM 4’s 16 GB is a more comfortable CPU-only experiment. Neither figure establishes useful generation speed.
Does Hostinger VPS include a GPU?
The KVM plan cards cited here list vCPU, RAM and NVMe storage and do not list a GPU. Choose a provider and plan that explicitly lists a compatible GPU if GPU inference is required.
What is the full Hostinger KVM 2 commitment?
The official plan page currently displays $8.79 per month equivalent and a $210.96 payment for 24 months, followed by renewal at $14.99 per month for another 24-month term. Check the checkout total because offers and taxes can change.