What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
The installation command depends on which “Spark” you have. On a Windows RTX Spark, run NVIDIA’s PowerShell installer; Ollama starts in the background and exposes its local API at http://localhost:11434. On a Linux-based NVIDIA DGX Spark, install the shell package so Ollama runs as a system service and detects the Blackwell GPU. Then pull a model, verify that it is present, and run it. Downloads stay on disk for reuse.
This guide covers both official workflows, model selection for chat, vision and agent tasks, Open WebUI, storage planning and the checks that matter when a model will not start.
Which RTX Spark setup do you have?
“RTX Spark” is used for more than one software setup in NVIDIA’s documentation. The two paths relevant to Ollama are:
- Windows RTX Spark: NVIDIA’s RTX Spark playbook uses PowerShell and an install script. Ollama launches automatically after installation and serves a local API.
- Linux DGX Spark: NVIDIA’s DGX Spark examples use a shell installer. Ollama runs as a system service and detects the Blackwell GPU.
Use only the command for your operating system. Do not paste the PowerShell command into a Linux terminal or the shell command into PowerShell.
#1 Best Overall
- Spacious Design: Measuring 21.1" wide and 14.1" deep, our lap desk comfortably fits most laptops up to 15.6". Extra room for accessories ensures convenience.
- Enhanced Functionality: Packed with handy features, including a 5x9" precision tracking mouse pad and a built-in phone slot for seamless work or video calls. Plus, enjoy ergonomic support with the integrated cushioned wrist rest.
- Cool Comfort: Enjoy a stable surface with our lap desk's dual bolster cushion, designed for comfort and airflow, keeping your lap cool during extended use.
- Durable Surface: Work with confidence on our lap desk's solid surface, featuring a sleek black carbon color, ensuring optimal air circulation to prevent your laptop from overheating.
- On-the-Go Convenience: With an integrated handle and lightweight design (2.8 lbs), our lap desk is portable for travel or moving around the house, offering flexibility in any space.
What Ollama can use on Spark
Ollama is a local model runtime and model manager. You download a model once, keep its files on the machine, and start it again without repeating the download. It is one of several backends NVIDIA presents for Spark; llama.cpp and LM Studio are separate choices. For the workflows here, Ollama is the most direct route to command-line model management and connections from local applications.
Ollama’s October 13, 2025 description of DGX Spark identifies the system as a GB10 Grace Blackwell machine with 128GB of memory and up to 1 petaFLOP of performance for prototyping and local language-model work. Those are platform figures from Ollama’s announcement, not a promise that every model will run at the same speed.
Ollama also reports that gpt-oss-120b fits in the 120GB of VRAM provided by the GB10 configuration it tested. Treat that as a tested/configuration claim: fitting in memory does not establish a universal token rate, latency or response quality for every Spark system.
Install Ollama on a Windows RTX Spark
NVIDIA’s Windows playbook installs Ollama with one PowerShell command. Open PowerShell under an account allowed to install software, then run:
irm https://ollama.com/install.ps1 | iexThe command downloads and executes Ollama’s installer from Ollama’s official endpoint. If your organization restricts script execution or outbound downloads, the command can be blocked by policy; use your normal approved software-installation process rather than bypassing that policy.
Confirm the Windows installation
- Wait for the installer to finish.
- Run
ollama --versionin PowerShell. - Confirm that a version string is returned.
NVIDIA says the Windows installation starts Ollama automatically in the background and serves its local API at http://localhost:11434. You do not need to run ollama serve or reopen the application for this workflow. A local client can use that address to send prompts after a model has been downloaded.
If PowerShell cannot find Ollama
- Close and reopen PowerShell so it reloads the executable path.
- Run
ollama --versionagain. - If it still fails, check whether your endpoint-security software blocked the installer and whether the installation completed under the account you are using.
- Before changing GPU settings, record the exact RTX Spark model, Windows version, Ollama version, driver status and available unified memory.
Install Ollama on a Linux DGX Spark
For the Linux-based DGX Spark workflow, open a terminal and run NVIDIA’s documented installer:
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 →Rank #2
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
curl -fsSL https://ollama.com/install.sh | shThen verify the command is available:
ollama --versionNVIDIA says this installation starts Ollama as a system service and detects the Blackwell GPU automatically. If the version command succeeds but a model later fails to load, investigate the service, driver and memory state rather than reinstalling the client repeatedly.
Initial DGX Spark setup time
NVIDIA notes that first-time DGX Spark setup can take as long as 10 minutes while updates install and the system joins the network. After setup, the machine can be used locally or accessed over the network. Allow that initialization to complete before diagnosing a model or service problem.
Download and run your first model
Installation and model download are separate operations. Installing Ollama gives you the runtime; ollama pull retrieves model files; ollama run starts an interactive session.
Windows RTX Spark example: Qwen3.8 27B
NVIDIA’s RTX Spark playbook uses the following sequence in PowerShell:
ollama pull qwen3.8:27b
ollama ls
ollama run qwen3.8:27bollama pull qwen3.8:27bdownloads the model.ollama lslists locally available models. You should see the exact tagqwen3.8:27b.ollama run qwen3.8:27bopens a prompt session.
NVIDIA says this Ollama tag already packages the Q4_K_M model and its Ollama parameters. You therefore do not need to copy a separate llama.cpp launch recipe or choose another quantization for this example.
Linux DGX Spark smoke test: Gemma 3 4B
For a small vision-language test, NVIDIA’s DGX Spark instructions use:
ollama pull gemma3:4b
ollama run gemma3:4bThis is a practical first check because the model is smaller than the larger general-purpose and agent recommendations. The same NVIDIA instructions list llama3.2-vision:11b as an alternative that may provide better quality at slower speed.
Rank #3
- Note: Not suitable for MacBooks released after 2023 or devices with a protruding front camera; Not applicable to full-screen or notch-style tempered glass screen protectors; Do not use on the rear camera of the phone.
- 💻 Why Do You Need a Webcam Cover Slide? — Safeguard your privacy by covering your webcam with our reliable webcam cover when not in use. Don't let anyone secretly watch you. Stay protected!
- ✅ Thin & Stylish — Enhance your laptop's functionality and aesthetics with our 0.027" ultra-thin webcam covers. Seamlessly close your laptop while adding a touch of sophistication.
- ✅ Fits Most Devices — Compatible with laptops, phones, tablets, desktops! Keep your privacy intact on Ap/ple, Mac/Book, iPh/one, iP/ad, H/P, L/novo, De/ll, Ac/er, As/us, Sa/msung devices.
- ✅ 365 Days Protection — Our upgraded 3.0 adhesive ensures a strong hold that won't damage your equipment. Experience reliable, long-term privacy protection day in and day out.
What happens after the first run
Model files remain on disk after the pull. Later runs use the local copy, so they do not require another download. Removing or replacing models is therefore a storage-management decision, not an installation requirement. Keep enough free space for every model and quantization you intend to retain; the exact disk consumption depends on those files and is not a single fixed value for all tags.
Choose a model by job, not just parameter count
Spark’s large unified-memory configuration makes models that would be awkward on a typical desktop practical, but the right choice still depends on the task, acceptable latency and available storage. NVIDIA’s official examples provide these starting points:
| Use case | Model tag | What the official guidance says | Trade-off to expect |
|---|---|---|---|
| General chat and first Windows test | qwen3.8:27b |
The RTX Spark playbook uses it and includes the Q4_K_M packaging and Ollama parameters. | Larger than a small smoke-test model; keep sufficient disk space and unified memory available. |
| Lightweight vision-language smoke test | gemma3:4b |
Recommended in NVIDIA’s DGX Spark VLM instructions. | Smaller and convenient for validation, but not the quality target for every visual or reasoning task. |
| Higher-quality vision-language alternative | llama3.2-vision:11b |
Listed by NVIDIA as an alternative that may provide better quality. | NVIDIA notes it may run more slowly than the 4B smoke-test option. |
| Agent-oriented work in Open WebUI | qwen3.6:35b-a3b |
NVIDIA lists this tag as its DGX Spark recommendation for the agent workflow. | Use it when agent behavior is the goal; it is a larger download than the 4B test model. |
Model names and tags are exact identifiers. Copy them exactly, including punctuation and the colon. A missing tag or a typo produces a “model not found” result even when Ollama itself is installed correctly.
Run Ollama through Open WebUI
Open WebUI is optional. Use it when you want a browser-based chat interface instead of an interactive terminal. NVIDIA describes it as a self-hosted interface in which models run locally on the GPU. The agent-ready workflow uses Open WebUI’s integrated Ollama stack to download and run qwen3.6:35b-a3b.
When Open WebUI is the better interface
- Choose the CLI when you are validating installation, scripting prompts or inspecting exactly which model tag is installed.
- Choose Open WebUI when several users or applications need a browser chat experience on the Spark host or its network.
- Use the integrated Ollama stack in NVIDIA’s documented agent workflow when you want the interface to handle the recommended model download and launch.
Keep the interface and runtime distinction clear: Open WebUI is the front end; Ollama remains the local model service. A browser page cannot compensate for an unavailable model, a stopped service or insufficient memory.
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 reinstallCrashes, 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 minuteConnect local applications to Ollama
On the Windows workflow, NVIDIA documents the local API at http://localhost:11434. Applications running on that same machine can target that endpoint after Ollama has started. Because localhost refers to the local computer, an application on another device needs a deliberately configured network-access path rather than assuming the address will resolve remotely.
For a basic connectivity check on Windows, first run ollama --version, then confirm that the model appears in ollama ls and start it with ollama run qwen3.8:27b. These checks separate three common failure classes: missing executable, missing model files and a model that cannot initialize.
Rank #4
- Anti-Slip Surface - Transform your laptop into a mobile workstation with the AboveTEK portable laptop lap desk. The anti-slip surface provides a strong grip for laptops up to 15.6 inches(Diagonal), while the double rubber strip on the bottom ensures a stable display or typing experience on your lap, couch, or bed.
- Retractable Mouse Pad - Retractable laptop mouse pad extends on both directions for the left/right handed with elevation along the edges for stopping mouse from falling off. The size of laptop tray is 14" X 9.7" and the size of mouse pad is 7.4" X 6.1".
- Effective Heat Shield - The effective heat shield made of sturdy and thick material protects your laptop from overheating. Prioritizes your comfort and safety, an ideal lap pad or board for working anywhere.
- EASY to Carry and Store - With an ergonomic and simplistic design, the lap desk is portable to store in a backpack. Only 15" in size, 2.2 lb of weight and with slim 0.6 inch thickness, it is ready to be easily carried around.
- Widely Applicable - The smooth platform accommodates laptops and tablets up to 15.6 inches(Diagonal), making it a versatile accessory and one of the best gifts for mom, dad, students and professionals. Perfect for use as a laptop bed tray or tablet holder anywhere at home, library, or park.
Memory, quantization and storage decisions
Unified memory is not the same as guaranteed speed
DGX Spark’s 128GB memory is a major reason larger local models are viable. Ollama’s report that gpt-oss-120b fits in 120GB of GB10 VRAM demonstrates a particular tested configuration, not a universal performance guarantee. Context length, model quantization, concurrent requests and other processes affect whether a workload is comfortable in practice.
Why quantization matters
Quantization reduces model storage and memory requirements, usually with a quality and sometimes a speed trade-off. The Windows example avoids a separate decision because NVIDIA says qwen3.8:27b already includes its Q4_K_M model packaging and Ollama parameters. For other tags, inspect the model’s published variant before planning capacity; do not assume that two tags with similar parameter counts occupy the same space.
Free tools Windows power users keep installed
One-click scans. No signup required.
Plan disk capacity before pulling several models
- List the models you actually expect to use, rather than downloading every available size.
- Leave room for the operating system, Ollama metadata and future updates.
- Remember that keeping both a small test model and a larger production model consumes space for both.
- An external SSD is an optional, configuration-dependent accessory, not a requirement established by NVIDIA’s Spark instructions.
Troubleshoot in the right order
When a model fails, check the environment before changing model commands. NVIDIA’s release notes identify CUDA Toolkit 13.0.2 and warn that GB10-based partner systems may not receive updates simultaneously. That means two Spark-branded systems can temporarily have different software states.
1. Confirm the platform and software versions
- Write down whether the machine is Windows RTX Spark or Linux DGX Spark.
- Run
ollama --version. - Check the GPU driver and CUDA installation reported by your system.
- Confirm the exact model tag and whether it appears in
ollama ls.
2. Separate download failures from load failures
If ollama pull cannot finish, investigate network access, disk space and the exact tag. If the pull succeeds but ollama run fails, investigate available unified memory, driver health and whether another process is using the GPU. Repeating the download will not fix a runtime or memory problem.
3. Check the service state on Linux
Linux DGX Spark is documented to run Ollama as a system service. If the command is installed but requests fail, inspect the service status and its logs using your distribution’s normal service-management tools. Confirm that the Blackwell GPU is visible to the driver before changing model settings.
4. Check the Windows background process
On Windows, Ollama is expected to start automatically in the background. If ollama --version works but the local API is unavailable, check whether the Ollama process was blocked or stopped and whether local security software is preventing access to http://localhost:11434. NVIDIA’s workflow does not require a separate ollama serve command.
5. Allow for first-boot updates
On a newly provisioned DGX Spark, wait through the update and network-join period, which NVIDIA says can take up to 10 minutes. Troubleshooting before that process finishes can make a normal initialization look like a model failure.
Best Value
- Spacious Design: Measuring 21.1" wide and 12" deep, our lap desk comfortably fits most laptops up to 15.6". Extra room for accessories ensures convenience.
- Enhanced Functionality: Packed with handy features, including a 5x9" precision tracking mouse pad and a built-in phone slot for seamless work or video calls. Plus, enjoy laptop support with the integrated device ledge.
- Cool Comfort: Enjoy a stable surface with our lap desk's dual bolster cushion, designed for comfort and airflow, keeping your lap cool during extended use.
- Durable Surface: Work with confidence on our lap desk's solid surface, featuring a blush pink color, ensuring optimal air circulation to prevent your laptop from overheating.
- On-the-Go Convenience: With an integrated handle and lightweight design (2.14 lbs), our lap desk is portable for travel or moving around the house, offering flexibility in any space.
Windows versus Linux: command reference
| Task | Windows RTX Spark | Linux DGX Spark |
|---|---|---|
| Install | irm https://ollama.com/install.ps1 | iex |
curl -fsSL https://ollama.com/install.sh | sh |
| Verify | ollama --version |
ollama --version |
| Service behavior | Starts automatically in the background; local API at http://localhost:11434. |
Starts as a system service and detects the Blackwell GPU. |
| Official first model example | qwen3.8:27b |
gemma3:4b for a lightweight VLM smoke test. |
| Model verification | ollama ls should show the pulled tag. |
Use ollama ls to confirm the local model before running it. |
A practical installation checklist
- Identify the operating system: Windows RTX Spark or Linux DGX Spark.
- Complete first-boot updates and network setup on a new DGX Spark; allow up to 10 minutes.
- Install Ollama with the matching command.
- Run
ollama --version. - Pull one appropriately sized model.
- Run
ollama lsand verify the exact tag. - Start it with
ollama run <model-tag>. - Keep the downloaded files and monitor free disk space before adding more models.
- Add Open WebUI only if you need a browser interface or the documented agent workflow.
- If anything fails, record the Spark variant, OS, Ollama version, driver/CUDA state, model tag and available unified memory before troubleshooting.
What to expect from local operation
Local execution keeps prompts and model inference on the Spark machine rather than requiring a hosted model endpoint. The practical benefit is control over the runtime and repeatable access after the initial download. The trade-offs are local disk consumption, responsibility for updates and the need to match model size and workload to the machine’s available memory.
For a quick proof that the stack works, use the smallest official example that matches your platform: qwen3.8:27b on the Windows playbook or gemma3:4b on the Linux DGX Spark VLM instructions. Move to llama3.2-vision:11b when visual quality matters more than speed, and use NVIDIA’s qwen3.6:35b-a3b recommendation when the goal is agent work in Open WebUI.
Frequently Asked Questions
Do I need to run ollama serve after installing Ollama on a Windows RTX Spark?
No. NVIDIA’s Windows playbook says Ollama starts automatically in the background and serves its local API at http://localhost:11434.
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 →Can I use the same Ollama command on Windows RTX Spark and Linux DGX Spark?
No. Windows uses the PowerShell installer at https://ollama.com/install.ps1; Linux uses the shell installer at https://ollama.com/install.sh.
Which model should I use first for a vision test?
NVIDIA’s DGX Spark VLM instructions use gemma3:4b as the lightweight smoke test and list llama3.2-vision:11b as a potentially higher-quality but slower alternative.
Will pulling a model download it every time I run it?
No. Ollama keeps downloaded model files on disk, so later runs reuse the local copy.
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.
Recommended Free Tools
