Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →The fastest supported installation on a Linux VPS is the Ollama installer: run curl -fsSL https://ollama.com/install.sh | sh, then verify it with ollama -v. For a server that must keep running after you disconnect SSH or reboot, run Ollama through its systemd service. The default API listens only on 127.0.0.1:11434; change that deliberately when another machine needs access.
Antihost offers Linux VPS plans with full root access, unlimited traffic on a 1 Gbps port and a 14-day free trial.
This guide covers Ubuntu and Debian-style VPSs, manual x86-64 and ARM64 installs, CPU-only and GPU-backed instances, persistent services, private API exposure, upgrades, logs, sizing, and complete removal.
As an Amazon Associate I earn from qualifying purchases.
Before you install: choose the VPS and access model
Ollama can run on a CPU-only VPS, so a GPU is not a prerequisite. NVIDIA CUDA and AMD ROCm are optional acceleration paths. The right VPS depends on the model you intend to load, its context size, the number of simultaneous requests, and whether inference runs on the CPU or an accelerator. The authoritative installation documentation does not publish one universal RAM, storage, CPU, or VRAM minimum for every model.
Confirm architecture and operating-system access
You need a 64-bit Linux VPS with administrative access (or an account that can use sudo). Check the architecture before choosing a manual archive:
#1 Best Overall
- AI Performance: 767 AI TOPS
- OC mode: 2632 MHz (OC mode)/ 2602 MHz (Default mode)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- A 2.5-slot design maximizes compatibility and cooling efficiency for superior performance in small chassis
uname -mx86_64corresponds to the x86-64 Linux archive.aarch64corresponds to the ARM64 Linux archive.
Make sure the VPS has enough free disk space for the operating system, Ollama itself, downloaded models, temporary files, and logs. Model storage can be substantially larger than the executable, and concurrent or larger-context workloads consume additional memory.
Decide where clients will connect
| Access arrangement | How it works | Security and operations |
|---|---|---|
| Localhost only | Ollama remains on 127.0.0.1:11434; applications on the same VPS call it locally. |
Smallest exposure surface and simplest setup. |
| Private interface or private network | Set OLLAMA_HOST to a private address or suitable bind value, then allow only trusted network sources. |
Useful for an application server on the same private network; requires firewall and routing discipline. |
| Reverse proxy | Keep Ollama behind a proxy that handles TLS, authentication, and access controls. | More components to maintain, but safer for controlled remote access than exposing an unauthenticated service directly. |
Install Ollama with the official Linux installer
For most Ubuntu and Debian VPSs, the one-line installer is the supported fast path. It detects the host architecture and available accelerator environment.
- Update package metadata and install the basic tools normally present on a minimal VPS. On Debian or Ubuntu, for example:
sudo apt update sudo apt install -y curl - Run the official installer:
curl -fsSL https://ollama.com/install.sh | sh - Verify that the executable is available:
ollama -v
A successful version response confirms that the command is installed and discoverable in your shell. If the shell cannot find ollama, start a new SSH session or check the installer output and your executable path before proceeding.
What the installer configures
When systemd is available, the installer creates an ollama service account when necessary, adds it to the render and video groups when those groups exist, writes /etc/systemd/system/ollama.service, and enables the service. The documented unit runs /usr/bin/ollama serve as user and group ollama, waits for network-online.target, restarts automatically, and uses a three-second restart delay.
The installer reports the local endpoint exactly as: “The Ollama API is now available at 127.0.0.1:11434.” That loopback binding means a remote computer cannot call the API until you explicitly configure a different host binding and the relevant network controls.
Run Ollama persistently with systemd
A systemd service is preferable to leaving ollama serve in an SSH terminal. It continues after your SSH connection closes, starts again after a reboot, and gives you standard status and log commands.
Recommended Free Tools
Enable, start, and inspect the service
- Ask systemd to reread unit files:
sudo systemctl daemon-reload - Enable Ollama at boot:
sudo systemctl enable ollama - Start it now:
sudo systemctl start ollama - Check its current state:
sudo systemctl status ollama - Read the service journal, including the most recent entries:
sudo journalctl -e -u ollama
active (running) in the status output indicates that systemd has a running process. The journal is the first place to look for permission errors, missing libraries, driver detection failures, bind-address mistakes, or model-loading problems.
Rank #2
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Verify the local API without downloading a model
From the VPS itself, make a basic HTTP request to the local listener:
curl http://127.0.0.1:11434/api/tagsThe endpoint should return a JSON response listing locally available models (which may be an empty list on a fresh installation). A connection refusal usually means the service is stopped or failed during startup; inspect systemctl status ollama and journalctl -e -u ollama before changing configuration.
Manual installation from a Linux archive
Manual installation is useful when you need a repeatable artifact, want explicit architecture control, or manage software through your own provisioning system. The Linux guide provides separate archives named ollama-linux-amd64.tar.zst for x86-64 and ollama-linux-arm64.tar.zst for ARM64. Select the archive that matches uname -m; do not install an x86-64 binary on an ARM64 VPS.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteWhen manual installation is the better choice
| Consideration | One-line installer | Manual archive |
|---|---|---|
| Repeatability | Simple to run, but behavior follows the current installer. | You can retain and deploy a specific archive through your own automation. |
| Architecture clarity | Detected automatically. | You select amd64 or ARM64 explicitly. |
| Package control | Convenient updates by rerunning the installer. | More control over which artifact is installed and when it changes. |
| Upgrade effort | Run the installer again. | Download, replace, and validate the selected archive yourself, then restart the service if applicable. |
For an AMD GPU, the guide calls for an additional ROCm archive in addition to the standard Linux package. Treat this as a separate accelerator installation path rather than assuming the generic archive contains every AMD component.
Use an NVIDIA or AMD GPU
CPU-only inference works without a GPU. A GPU-backed VPS can improve throughput or make larger models practical, but the host must provide a compatible device, driver stack, and enough memory for the selected workload.
NVIDIA-backed VPS
- Choose a VPS plan that actually passes through an NVIDIA device to the guest. A plan advertised as “GPU” is not sufficient unless the device is visible inside your VPS.
- Install the cloud provider’s compatible NVIDIA/CUDA driver separately from Ollama.
- Verify the driver and device from the VPS:
nvidia-smi - Install Ollama using the standard Linux installer, then inspect the service journal while loading a model.
If nvidia-smi fails, fix the provider’s device passthrough or driver installation before troubleshooting Ollama. Driver version compatibility, kernel updates, and provider support are ongoing maintenance considerations.
AMD-backed VPS
- Confirm that the provider exposes an AMD GPU to the guest.
- Install the current ROCm stack recommended for that operating system and kernel.
- Install the extra Ollama ROCm package described by the Linux guide.
- Review service logs while starting a model and verify that the expected device is detected.
The guide warns that older kernel AMDGPU support may not expose every ROCm feature and recommends the latest AMD Linux driver for best support. A device being visible at the kernel level therefore does not guarantee that every ROCm capability is available to Ollama.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
CPU-only deployment
For a CPU VPS, skip CUDA and ROCm packages. Start with a model appropriate to the available memory and test one request at a time. Increase concurrency or context size only after monitoring memory use and response times; these settings can change the resource requirements even when the executable and model remain unchanged.
Download and run a model
Ollama installs the runtime, not a model collection. Download a model only after the service is healthy and you have confirmed available disk and memory. The exact model command depends on the model name you select. A typical workflow is:
- Check that the service is running:
sudo systemctl status ollama - Use the Ollama command-line interface to fetch the model you have chosen:
ollama pull <model-name> - Start an interactive session when the download completes:
ollama run <model-name> - For an application, call the local HTTP API at
http://127.0.0.1:11434rather than keeping an interactive shell open.
Replace <model-name> with a real model identifier; the angle-bracket text is a shell placeholder, not a model name. Large models can take considerable time and storage to download, especially on a VPS with limited disk throughput.
Expose the API to another server safely
Do not change the bind address merely because a remote client cannot connect. The loopback default is a useful safety boundary. If another server must call Ollama, configure OLLAMA_HOST in the systemd service and restrict network access.
Set a service environment override
- Open a systemd drop-in for Ollama:
sudo systemctl edit ollama - In the editor, add an environment assignment using the private address or bind value appropriate for your network. For example:
[Service] Environment="OLLAMA_HOST=0.0.0.0:11434" - Save and close the editor.
- Reload systemd and restart Ollama:
sudo systemctl daemon-reload sudo systemctl restart ollama - Check that the service remained healthy:
sudo systemctl status ollama sudo journalctl -e -u ollama
0.0.0.0:11434 listens on all interfaces and should not be treated as an authentication mechanism. Prefer the VPS’s private interface when possible. If a reverse proxy is required, keep Ollama reachable only from that proxy and enforce authentication, TLS, and source restrictions there.
Test from the client server
From the authorized client, call the VPS’s private address:
curl http://<private-vps-address>:11434/api/tagsReplace the placeholder with the actual private address. If the local request succeeds but the remote request times out, inspect the VPS firewall, cloud security group, private-network route, and listening address. If the remote request is refused, inspect the service status and the bind value first.
Rank #4
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Exposure checklist
- Allow port 11434 only from the application server or private network, not from the entire public internet.
- Use a reverse proxy when you need TLS termination, authentication, rate limiting, or access logging.
- Do not assume the Ollama API itself supplies the authentication boundary you need.
- After every configuration change, test locally and then from the intended client.
- Keep the systemd override documented so upgrades and incident recovery do not erase your intended network policy.
Updates, version pinning, and rollback planning
Update to the current installer release
Rerun the official installer to update an existing installation:
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows 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 reinstallcurl -fsSL https://ollama.com/install.sh | shAfterward, verify the installed version and restart or inspect the service if your deployment requires it:
ollama -v
sudo systemctl status ollamaPin a release or test a pre-release
The Linux guide supports passing OLLAMA_VERSION=<version> to the installer. Its example pins version 0.5.7:
curl -fsSL https://ollama.com/install.sh | OLLAMA_VERSION=0.5.7 shUse a version that is valid for your deployment; the example is a documented syntax example, not a recommendation that every VPS should run that release. Pinning is useful when you need reproducible provisioning or want to test an upgrade separately from production.
Protect service configuration during upgrades
Keep custom settings such as OLLAMA_HOST in the systemd drop-in created by systemctl edit ollama, rather than editing the vendor unit directly. After an update, run systemctl daemon-reload, restart if needed, and confirm both the environment and listener with status output and a local API request.
Logs and common failure branches
The command is not found
Open a new login shell and run ollama -v again. If it still fails, inspect the installer output and verify that the binary was installed in the expected path. The documented service invokes /usr/bin/ollama, so a running systemd unit can still work even when a particular user’s interactive PATH is incomplete.
The service will not start
Run:
sudo systemctl status ollama
sudo journalctl -e -u ollamaLook for malformed drop-in syntax, permission failures, missing shared libraries, an occupied port, or an accelerator driver error. Correct one cause at a time, then restart the service and recheck the journal.
Best Value
- Powered by the NVIDIA Blackwell architecture and DLSS 4 OC mode: 2640MHz/Default mode: 2610MHz (Boost Clock)
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.125-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
Remote connections fail after changing the host
First test curl http://127.0.0.1:11434/api/tags on the VPS. If that fails, the service or its bind setting is wrong. If it succeeds, verify the private address, firewall rule, cloud security group, routing, and client-side network policy. A broad public bind can also create an unintended exposure, so narrow the address and source rules as soon as connectivity is confirmed.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →GPU is not being used
Check the provider’s device passthrough and run nvidia-smi for NVIDIA instances. For AMD, verify the ROCm installation and current driver support. Then inspect journalctl -e -u ollama. A CPU fallback may indicate a driver, permission, kernel, or unsupported-feature problem rather than an Ollama installation problem.
The model download or load fails
Check free disk space, available RAM or VRAM, network stability during the download, and the model’s resource demands. A model that fits on one machine may not fit on a smaller VPS, and increasing context size or concurrency can turn a previously successful load into an out-of-memory failure.
Remove Ollama completely
Uninstallation removes the service and downloaded model data, so export or back up anything you need first. The official Linux sequence is:
- Stop the service:
sudo systemctl stop ollama - Disable it at boot:
sudo systemctl disable ollama - Remove the systemd unit and binary according to the uninstall procedure for your installation.
- Delete downloaded models under
/usr/share/ollama. - Remove the
ollamaservice account and group.
After removal, confirm that the unit no longer appears in systemctl status ollama and that the model directory is gone if complete data removal was your goal. If you installed a GPU driver or ROCm/CUDA stack for other workloads, treat those packages separately; removing Ollama does not automatically mean those provider components should be removed.
Choose a deployment pattern
| Pattern | Best fit | Main trade-off |
|---|---|---|
| CPU VPS, localhost API | Development, internal tools, low concurrency, and applications on the same host. | Lower infrastructure complexity, but throughput and model size are limited by CPU and memory. |
| NVIDIA GPU VPS | Higher throughput or models that are impractical on the available CPU. | Requires device passthrough and continuing CUDA driver maintenance; hourly cost is provider-dependent. |
| AMD GPU VPS | Hosts where ROCm is supported and the provider exposes a compatible AMD device. | ROCm and kernel/driver compatibility require careful validation. |
| Private-network API | A separate application server needs Ollama without public exposure. | Requires private routing, firewall rules, and a controlled bind address. |
| Reverse-proxied API | Multiple clients or internet-facing workflows that need TLS and authentication controls. | More operational components and monitoring than a localhost-only service. |
Practical readiness checklist
- Architecture:
uname -mmatches the selected package or installer detection. - Installation:
ollama -vreturns a version. - Persistence:
systemctl status ollamareports an active service. - Local API:
curl http://127.0.0.1:11434/api/tagsreturns JSON. - Storage: The VPS has room for the selected models and future downloads.
- Accelerator: NVIDIA hosts pass
nvidia-smi; AMD hosts have the required ROCm stack and driver support. - Network: Any non-local bind is limited by private routing, firewall rules, and authentication controls.
- Operations: You know where to find logs and how to restart or pin the service version.
Frequently Asked Questions
Can Ollama run on a Linux VPS without a GPU?
Yes. CPU-only operation does not require CUDA or ROCm. Select a model and concurrency level that fit the VPS’s available memory and CPU capacity.
What is Ollama’s default API address?
The installer reports 127.0.0.1:11434. It is local to the VPS until you configure OLLAMA_HOST for a private interface or another deliberate access method.
How do I keep Ollama running after I close SSH?
Use the systemd service: enable it with sudo systemctl enable ollama and start it with sudo systemctl start ollama. Check it with sudo systemctl status ollama.
How do I check Ollama service logs?
Run sudo journalctl -e -u ollama. The journal shows startup, restart, bind, permission, driver, and model-loading errors.
Free tools Windows power users keep installed
One-click scans. No signup required.
How do I install a specific Ollama version?
Pass a version to the installer, for example curl -fsSL https://ollama.com/install.sh | OLLAMA_VERSION=0.5.7 sh. Use a valid release identifier for your deployment.
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.
