Recommended Free Tools
Antihost VPS is listed first here as a Linux VPS option with AMD Ryzen 9 9950X3D CPUs, NVMe storage, full root access and up to 64 GB RAM. For most Ollama users, 16 GB RAM, fast AMD EPYC-class CPUs and at least 100 GB of NVMe storage provide enough headroom for quantized 7B–8B models and a practical path to 13B models. An 8 GB machine is the minimum sensible starting point; 70B models generally need 48–64 GB or more and are not a good fit for an ordinary shared VPS. Choose a GPU server only when low latency, long context or larger models justifies the higher and less predictable cost.
What is the best VPS size for Ollama?
Ollama runs the model weights, runtime and context window in memory. On a CPU-only VPS, system RAM is the limiting resource: when the model and its context do not fit, Linux starts swapping, token generation becomes extremely slow, and the process can be killed by the out-of-memory manager.
As an Amazon Associate I earn from qualifying purchases.
| Model class | Typical quantized RAM guidance | Recommended VPS | Practical expectation |
|---|---|---|---|
| 3.8B (for example, a small Phi model) | About 4 GB in one vendor roundup | 8 GB RAM, 2–4 modern vCPUs | Good for experiments and lightweight chat; leave room for the operating system. |
| 7B–8B | 8 GB minimum in the Ollama model library; 8–10 GB for Q4 builds in another roundup | 16 GB RAM, 4–8 modern vCPUs | The sensible general-purpose tier for an always-on assistant. |
| 13B–14B | 16 GB in the model-library guidance; roughly 16–18 GB for Q4 in another roundup | 24–32 GB RAM, 8–16 vCPUs | Usable on CPU, but generation is slower and context length consumes additional memory. |
| 70B | 48 GB or more in a roundup; 64 GB in the Ollama model-library guidance. A community Q4 estimate is about 40 GB. | 64 GB or more, preferably dedicated high-memory hardware | Usually impractical on a standard VPS because of cost, contention and CPU latency. |
These are planning figures rather than a formal Ollama minimum. Ollama’s official documentation does not publish one universal RAM floor. Quantization level, context length, batch size, operating-system overhead and concurrent requests all change the actual requirement. Reserve at least 2–4 GB for Linux, Ollama and monitoring instead of assigning every gigabyte to the model.
CPU-only Ollama: when a VPS works without a GPU
Yes, a VPS without a GPU can run quantized 7B and 8B models. CPU inference is attractive when requests are occasional, the service is asynchronous, or the goal is private automation rather than an interactive, low-latency chat interface. It also avoids GPU-driver maintenance and the scarcity pricing common with accelerator instances.
#1 Best Overall
- 12th Intel Alder Lake N95 Processor – The GMKtec G3 S Mini PC is powered by the 12th Gen Intel N95 processor with 4 cores, 4 threads, 6MB cache and a burst frequency up to 3.4GHz. Compared with N100/N5105/N5100/N5095, the N95 delivers up to 36% overall performance improvement. Perfect for routine tasks, office work, and home entertainment, this compact mini desktop is more convenient than traditional bulky PCs.
- 8GB RAM & 256GB SSD Storage – Pre-installed with 8GB DDR4 memory and a fast 256GB M.2 2242 SSD, the G3 S mini desktop offers quicker startup, smoother multitasking, and faster file transfers. Enjoy seamless performance whether you’re working on multiple applications, browsing, or streaming content.
- Rich Interfaces & Connectivity – The G3 S mini computer comes equipped with USB 3.2 (up to 10Gbps), dual HDMI 2.0 (4K@60Hz), and a 3.5mm audio jack. With support for WiFi 5, Bluetooth 5.0, and Gigabit Ethernet (RJ45 1000MbE), it connects easily with monitors, projectors, printers, office equipment, and other peripherals, making it versatile for both home and business use.
- Dual 4K Display Support – Featuring upgraded Intel UHD Graphics (up to 1000MHz), the G3 S supports 4K video playback and AV1 decoding for a smooth viewing experience. With dual HDMI outputs, you can connect two 4K@60Hz displays simultaneously, enabling efficient multitasking for work and entertainment.
- GMKTEC WARRANTY - GMKtec offers a 3-year limited warranty (1 year replacement + 2 years parts replacement) for each mini PC, starting from the date of the purchase effective on all sales starting Oct. 2026. All defects due to design and workmanship are covered. With a professional after sales team always ready to attend to your needs, you can simply relax and enjoy your mini PC
What performance to expect
One roundup quotes 20–40 tokens per second for 7B–8B CPU inference, but the same source set describes that range as optimistic. Results vary with the CPU generation, number of physical cores, memory bandwidth, quantization, thread count and whether other tenants share the host. Treat any advertised number as a hypothesis: run your own prompt and measure time to first token and sustained tokens per second on the exact plan.
- Interactive chat: prioritize faster single-core and memory performance, even if that means fewer total vCPUs.
- Batch jobs or background agents: more cores can improve throughput when you run several independent generations.
- Multiple users: size for concurrent memory use first, then CPU. Two simultaneous 8B requests can need substantially more RAM than one request.
A VPS becomes a poor choice when you need instant responses from a 30B–70B model, many concurrent users, or long context windows. At that point, compare a dedicated GPU server or GPU cloud instance instead of simply buying a larger shared CPU plan.
How much storage does an Ollama VPS need?
Model files are large, and Ollama keeps downloaded models on disk even after they are loaded into RAM. Community guidance commonly uses about 50 GB of free space as a baseline. That is enough for one or two smaller models plus the operating system, but it leaves little room for updates, logs or a second 13B model.
- 8 GB RAM plan: target at least 60–80 GB NVMe so a small model, swap file and updates do not fill the disk.
- 16 GB plan: 100–160 GB NVMe is more comfortable for several 7B–8B variants and one 13B model.
- 32 GB or larger: 200 GB or more avoids constant model deletion when testing different quantizations.
NVMe improves model loading and startup time, but it does not replace RAM. Increasing disk size will not make a model that does not fit in memory usable. Check whether a provider’s advertised capacity is NVMe and whether snapshots or backups consume additional billable storage.
Context length and VRAM planning
Ollama’s default context setting is tied to available VRAM when a GPU is used: under 24 GiB, the default is 4,096 tokens; from 24 to 48 GiB, 32,768 tokens; and at 48 GiB or more, 262,144 tokens. These are defaults, not guarantees that every model can process that much context. Longer context increases memory use and can reduce speed.
Rank #2
- [15W Ryzen 7 Agentic PC for Everyday Workflows] Powered by the AMD Ryzen 7 7730U processor (8 Cores, 16 Threads), the GEEKOM A5 is built for sustained productivity. It doubles as your cloud-native Agentic AI assistant, seamlessly hosting cloud AI tasks, automating office workflows, and handling intelligent document summarization without complex local deployment. Smoothly manage Microsoft Office, dozens of browser tabs, heavy Excel spreadsheets, and remote learning throughout your workday.
- [Smart Value Now, Expandable for Tomorrow] Equipped with 16GB RAM and a fast 256GB PCIe NVMe SSD for snappy daily performance, the A5 offers incredible value. Need more space later? It features dual-slot DDR4 RAM (upgradable to 64GB) and supports an M.2 SSD up to 4TB. With an extra M.2 2242 slot and 2.5" HDD bay for up to 10TB total storage, you get the flexibility to scale your storage seamlessly as your needs grow, beating soldered LPDDR solutions.
- [Multi-Display Connectivity for Maximum Productivity] Create a complete workstation with support for up to four displays through Dual HDMI and Dual USB-C ports, including up to 8K output via USB-C. Stay connected with Wi-Fi 6, Bluetooth 5.4, a 2.5GbE LAN port, SD card reader, and multiple USB ports for fast networking, efficient multitasking, and seamless connectivity across all your devices.
- [Built to Stay Cool, Quiet & Reliable] More than fast, the GEEKOM A5 is built to last. A reinforced one-piece all-metal internal frame enhances structural strength, while the upgraded IceBlast 3.0 cooling system improves cooling efficiency by up to 42% with up to 35% greater airflow for quieter operation. Backed by 339 reliability tests and a 72-hour full-load aging test, it's engineered for dependable long-term performance.
- 🏢[Business-Ready, Compact & Efficient] Pre-installed OS, the GEEKOM A5 supports Wake-on-LAN, Scheduled Power On, and Group Policy, making deployment and remote management simple for businesses. Its ultra-compact 0.6L design fits neatly behind monitors or into space-limited workstations while delivering excellent power efficiency for home offices, front desks, and commercial environments.
On a CPU-only VPS, the same practical rule applies: start with a modest context, confirm memory usage, and increase it only when the workload requires it. A 16 GB machine running an 8B model may be comfortable at a short context but run out of memory when a large context, tool output and conversation history are included.
Provider comparison for an Ollama VPS
Provider figures below come from 2026 roundups and are not independently verified on provider pages. Prices and plan names can change, promotional rates may not renew at the same amount, and shared-CPU performance can differ between locations. Use the figures as a shortlist, then confirm the live offer, billing term, bandwidth policy and virtualization type before paying.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Provider and cited plan | Published specification | Why it is considered | Important qualification |
|---|---|---|---|
| Antihost VPS | Linux VPS with AMD Ryzen 9 9950X3D, NVMe storage, 1–16 vCPUs and 2–64 GB RAM | Full-root VPS for self-hosted Ollama with high-memory tiers | Self-managed; no cPanel or Windows. Confirm the current plan, region and billing term. |
| Contabo Cloud VPS 6 | 6 vCPU, 12 GB RAM, $7.20/month in one 2026 roundup | High RAM per dollar for 7B–8B models | Roundup says shared CPU is weaker than faster alternatives; price and availability were not checked against the provider. |
| Contabo Cloud VPS S | 8 GB RAM at about €6/month in another roundup | Low-cost entry point for a single small model | Different roundup and plan naming; verify current CPU, storage and renewal terms. |
| Hetzner CPX41 | 8 vCPU, 16 GB RAM, 160 GB NVMe, €14.99/month as listed in a roundup | Balanced memory, storage and faster AMD EPYC-class CPU performance | Listed price is not a live quote and may vary by location or billing conditions. |
| Hetzner CPX51 | 16 vCPU, 32 GB RAM, 240 GB NVMe, €19.99/month as listed in a roundup | Headroom for 13B models, multiple model files and concurrent jobs | Confirm current availability and whether the plan is shared or dedicated. |
| Netcup | Roundup schemas show roughly €13–€32/month for 10–16 vCPU and 16–48 GB RAM | Several high-memory tiers for users who have outgrown 16 GB | Exact plan, disk and renewal details were not established. |
Best value for a small personal assistant
A 12–16 GB plan is the value sweet spot. The lower-priced 8 GB tier can run one 7B–8B quantized model, but has little room for a large context, a database, a web UI or monitoring. Spend more on CPU only when you have measured that generation speed, rather than RAM, is your bottleneck.
Best performance among CPU VPS options
The roundups favor Hetzner’s AMD EPYC-based CPX plans for CPU performance. That recommendation is directional, not a benchmark: token-speed claims conflict, and no standardized test was supplied. Select a nearby region, run the same prompt and record tokens per second before committing to a long billing term.
When the cheapest RAM-per-dollar plan wins
Contabo’s cited plans offer more memory for less money, which matters when the model barely fits. They are sensible for offline or asynchronous work where a slower response is acceptable. Avoid treating a low monthly price as a performance guarantee; shared CPU contention can dominate your experience.
Rank #3
- 【AMD Ryzen 3 5300U CPU: Outperforms N150 & 3500U】 BOSGAME E5 mini PC is powered by the TSMC 7nm FinFET architecture AMD Ryzen 3 5300U processor (4 Cores, 8 Threads, up to 3.8GHz boost, 6MB total cache). Compared to low-end Intel N150 or 3500U chips which only have 4 single threads and throttle under load, the 5300U delivers over 30% faster multi-core speed. Run 30+ browser tabs, large Excel sheets, and Zoom meetings simultaneously without system lag.
- 【8GB DDR4 RAM & 256GB NVMe SSD Storage】 Installed with high-speed 8GB DDR4 dual-channel memory and a fast 256GB M.2 2280 SSD, eliminating slow boot times and application loading delays. To accommodate growing data requirements, the upgradeable hardware design features dual SODIMM slots that allow you to expand memory up to 64GB RAM, ensuring smooth operation during heavy multitasking.
- 【High-Capacity Dual M.2 SSD Storage Expansion】 Never worry about running out of space for your business files. In addition to the pre-installed 256GB system drive, the motherboard houses an extra empty internal M.2 2280 NVMe PCIe 3.0 slot. This allows you to easily add a second solid-state drive for up to an additional 2TB of storage capacity (upgrades not included) without needing to remove or reinstall the original operating system.
- 【Radeon 6-Core Graphics & Triple 4K Displays】 Integrated with official AMD Radeon Graphics (6 Graphics Cores, 1500 MHz frequency) for casual gaming, photo editing, and crisp 4K media decoding. Featuring 1x HDMI 2.0 port, 1x DisplayPort, and 1x Full-Function Type-C port, the E5 outputs true 4K@60Hz resolution to three monitors at once. This multi-screen setup eliminates constant window-switching for traders, programmers, and office workers.
- 【Dual 2.5GbE LAN Ports for Advanced Networking】 Experience fast wired network transmission speeds up to 2500Mbps without lagging or buffering. The integration of dual 2.5 Gigabit Ethernet ports (powered by Realtek RTL8125 controller) makes this compact computer an exceptional hardware choice for tech enthusiasts. Easily configure it into software routers, hardware firewalls (pfSense, OpnSense), home NAS servers, or local homelabs.
GPU VPS requirements for Ollama
GPU acceleration changes the decision from “which ordinary VPS has enough RAM?” to “which host exposes a compatible device and driver?” Standard VPS products generally have no GPU. You need a dedicated GPU server or a GPU-cloud instance with passthrough or a supported virtual GPU.
NVIDIA on Linux
Ollama’s official documentation specifies NVIDIA compute capability 5.0 or newer with driver 550 or newer. GPUs with compute capability 5.0–6.2 require driver 570 or newer. Confirm the exact GPU model and driver before provisioning; a generic “CUDA” label is not enough.
AMD on Linux
AMD acceleration requires ROCm version 7. Some distributions also require the Ollama user to belong to the render group so the service can access the device.
Other accelerators
Vulkan support is available for other GPUs, but it may require additional components and root privileges or capabilities for VRAM visibility. Treat this path as an engineering project rather than a plug-and-play VPS feature.
GPU cost decision
A roundup gives a rough self-hosted GPU range of $500–$1,500 per month, but that figure is indicative and was not source-checked. No current, apples-to-apples GPU VPS price survey is established here. Get an hourly and monthly quote, ask whether the GPU is dedicated, and calculate the cost from your expected uptime before choosing acceleration.
Rank #4
- This Certified Refurbished product is tested and certified to look and work like new. The refurbishing process includes functionality testing, basic cleaning, inspection, and repackaging. The product ships with all relevant accessories, a minimum 90-day warranty, and may arrive in a generic box. Only select sellers who maintain a high-performance bar may offer Certified Refurbished products on Amazon.com.
- Dell OptiPlex 7050 Micro Computer, Intel Quad Core i5-6500T up to 3.1GHz, 16G DDR4, 256G SSD.
- Includes: USB Keyboard & Mouse, Microsoft office 30 days free trail.
- Ports: 1 x RJ-45, 1 x HDMI, 1 x DP, 6 x USB 3.0.
- 4K Support: Support 4K (3840x2160) Dual display, makes it easy to connect two monitors at the same time, and you can expand working Windows, mirror content, or expand a single window across multiple monitors.
Deploy Ollama on a Linux VPS
The following sequence keeps the installation reproducible without assuming a particular provider dashboard. Use a current supported Linux distribution and an account with sudo access.
- Choose the machine: start with 16 GB RAM, 100 GB or more NVMe and at least four modern vCPUs for an 8B model. Select 32 GB RAM for a 13B model or multiple concurrent jobs.
- Update the operating system: run your distribution’s package update and reboot if the kernel or security libraries changed.
- Install Ollama: follow the current Linux installation instructions from the official Ollama documentation, then verify the binary with
ollama --version. - Start the service: run
ollama serveduring initial testing. For production, enable the vendor-provided system service so it restarts after a reboot. - Download one model: use
ollama pull <model-tag>, replacing the tag with the exact model and quantization you selected. Check free disk space before pulling additional models. - Run a local test: execute
ollama run <model-tag>and send a fixed prompt. Record time to first token, sustained tokens per second and peak RAM. - Measure under your workload: repeat the test with your normal context length and tool output. A short benchmark prompt can hide memory pressure that appears in production.
- Expose an API cautiously: keep Ollama bound to localhost unless a reverse proxy, authentication and TLS protect it. Do not open the Ollama port directly to the public internet.
- Automate health checks: monitor RAM, swap, disk usage, load average and restart count. Alert before disk reaches 80% or available memory falls below the operating-system reserve.
Security and reliability checklist
- Use SSH keys, disable password login after verifying key access, and restrict SSH to your administration IP range where possible.
- Apply security updates and enable the provider firewall plus a host firewall. Permit only SSH, your reverse-proxy ports and required application traffic.
- Put an authentication layer in front of any remote Ollama API. The default local API is not an identity system.
- Run Ollama as a dedicated unprivileged user. Grant GPU or render-group access only when required.
- Use a persistent volume or tested backup for prompts, application data and configuration. A VPS snapshot is not a substitute for an independent backup.
- Set request timeouts and concurrency limits. Unlimited simultaneous generations can exhaust RAM even when one request is stable.
Troubleshooting common VPS failures
The model exits with an out-of-memory error
Check free RAM while the model is loaded, reduce context length, close other services and remove unused models from disk. If the model still does not fit with a safe operating-system reserve, move to the next RAM tier; adding vCPUs will not solve a memory failure.
Generation is extremely slow
Confirm that the VPS actually has the advertised CPU, inspect load and steal time, and test at a quiet period. Shared-host contention, an oversized context or swap usage are common causes. Compare a second provider with the same model and prompt before concluding that Ollama itself is at fault.
The GPU is detected but unused
Verify the GPU model, compute capability, driver version and device permissions. NVIDIA systems below the documented driver requirement, AMD systems without ROCm 7, and Linux users missing render-group access can fall back to CPU or fail to initialize.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →The service works locally but not remotely
Check the service bind address, reverse-proxy route, firewall rules and TLS certificate. Keep the Ollama listener private and expose only the authenticated proxy endpoint.
Disk fills after testing models
List downloaded models, remove unused variants, rotate logs and increase the volume before the filesystem reaches capacity. Keep free space for temporary downloads and upgrades; a nearly full disk can make an otherwise healthy service fail.
A practical buying decision
- Pick the model first. For a 7B–8B quantized model, budget 16 GB RAM; for 13B–14B, budget 24–32 GB; for 70B, look beyond a standard VPS.
- Reserve storage. Start at 100 GB NVMe for a multi-model setup and use 50 GB only as a bare minimum for a single small model.
- Choose CPU based on latency. Favor faster EPYC-class performance for interactive use; favor more cores for parallel batch jobs.
- Decide whether you truly need a GPU. If your target is a 7B–8B model and requests are occasional, CPU-only is simpler and cheaper. Move to GPU hardware for sustained interactive speed, large models or long contexts.
- Validate the live offer. Recheck price, renewal rate, region, storage type, bandwidth, backup cost, virtualization and whether resources are shared.
- Run a representative trial. Keep the plan only if it meets your measured latency and memory target with the context and concurrency you actually use.
Antihost VPS is listed first here: its Linux VPS plans use AMD Ryzen 9 9950X3D CPUs, NVMe storage, full root access and up to 64 GB RAM. For most readers, the defensible 2026 starting point is a 16 GB Linux VPS with 100–160 GB NVMe and a modern shared CPU, using a quantized 7B–8B model. Choose the lower-cost high-memory option when budget matters more than response speed; choose the faster 16–32 GB tier when interactive latency or multiple requests matters. Treat 70B CPU hosting and GPU pricing as separate, specialized decisions rather than ordinary VPS upgrades.
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.
