Hands-on impressions using a Snapdragon X Elite laptop

Local LLMs have exploded in popularity, and with new hardware like the Snapdragon X Elite, it’s finally practical to run powerful AI models entirely on your machine—fast, private, offline, and inexpensive.

In this guide, I’ll walk you through three of the most popular tools for running local models:

For each, I’ll cover:

  • How to install it

  • Minimum hardware requirements

  • Rough model availability

  • Pros and cons

  • My personal performance testing results

All tests were done on:

🧑‍💻 My Test Hardware

Microsoft Surface Laptop, 7th Edition

ComponentDetails
CPUSnapdragon X Elite (X1E80100), 12 cores @ 3.40 GHz
RAM32 GB
OSWindows 11 Home, Build 26200
GPU / AI AccelerationDirectX 12 / NPU support
NotesARM-based architecture

This hardware is extremely efficient for local inference—especially for optimized models.

1. Ollama

✔️ “The easiest way to run local LLMs… period.”

Ollama has become the go-to local model runner due to its simplicity. It's lightweight, command-line driven, and supports hundreds of models.

How to Install Ollama (Windows ARM)

  1. Go to: https://ollama.com/download

  2. Download the Windows ARM installer

  3. Run and follow the prompts

  4. Open PowerShell and test:

        ollama run llama3.1

That’s literally it.

Minimum Hardware Recommendations

ComponentMinimum
RAM8–16 GB
CPU4–8 cores
Disk4–20 GB free for models
GPUOptional — CPU-only supported

Ollama is extremely efficient on both Intel/AMD and ARM.

Model Availability

  • ~200+ public models

  • Many variants: Llama, Phi, Mistral, Gemma, Yi, Qwen, CodeGemma, etc.

  • Supports GGUF models

  • Can load custom local models

Pros

  • ✔ Ridiculously easy to install

  • ✔ Command-line interface is great for automation

  • ✔ Large community and ecosystem

  • ✔ Supports embeddings and system-level model management

  • ✔ Great for developers, scripting, and agents

Cons

  • ✖ No polished UI (unless you install third-party UIs)

  • ✖ Slower than LM Studio and Foundry in my tests

  • ✖ Some models load slower initially

My Testing Results (Snapdragon X Elite)

  • Easy to install and get running

  • Slower response time than LM Studio and Foundry

  • Stable and reliable, but not the fastest

  • Perfect for scripting, agents, and dev workflows


2. LM Studio

✔️ “The most polished local LLM experience.”

LM Studio is a desktop app with a great UI, a built-in model browser, and a fast runtime.

How to Install LM Studio

  1. Visit: https://lmstudio.ai/

  2. Download the Windows installer (ARM support is now available)

  3. Install and open the app

  4. Browse models and click Download + Run

LM Studio handles everything for you.

Minimum Hardware Recommendations

ComponentMinimum
    RAM     16 GB
    CPU      8 cores or better
    Disk    10–50 GB, depending on models
    GPU/NPU    Optional but beneficial

Model Availability

  • ~600+ models searchable in-app

  • HuggingFace integration

  • Auto-optimized versions for your architecture

  • Great model tagging and recommendations

Pros

  • ✔ Best UI out of the three

  • ✔ Very fast responses on ARM/CPU

  • ✔ Easy for beginners

  • ✔ Built-in chat, system prompts, memory, logs

  • ✔ Auto-downloads correct GGUF format

Cons

  • ✖ Larger install footprint

  • ✖ More RAM-intensive

  • ✖ Less automated for scripting/agents

My Testing Results (Snapdragon X Elite)

  • Easy to install and use

  • More polished interface than Ollama

  • Fast responding — the fastest of all three

  • Feels like ChatGPT running locally

  • Great candidate for daily use

3. Microsoft Foundry

✔️ “A lightning-fast, developer-focused local LLM runtime.”

Microsoft’s local LLM runtime (Foundry) is now one of the easiest and fastest ways to run models—especially on AI-enabled ARM hardware.

It’s command-line driven and built for developers.

How to Install Foundry (Windows PowerShell)

Open PowerShell and run:

    winget install Microsoft.Foundry

Then test:

    foundry chat run model phi-4

That’s it. It installs in seconds.

Minimum Hardware Recommendations

ComponentMinimum
RAM16 GB
CPU8 cores
GPU/NPUOptional, but highly optimized for NPU + ARM
Disk4+ GB

Model Availability

The list is smaller, but incredibly optimized.

Pros

  • ✔ Fastest local inference in my testing

  • ✔ Simple one-line installation

  • ✔ Native ARM + NPU acceleration

  • ✔ Command-line friendly

  • ✔ Small footprint

Cons

  • ✖ No GUI

  • ✖ More developer-centric

  • ✖ Smaller model catalog

My Testing Results (Snapdragon X Elite)

  • Simple to install via one PowerShell command
    Command-line based
    Extremely fast responses
    Best performance per watt on ARM/NPU
    Great for automation and agents

Final Comparison Table

Feature Ollama LM Studio Foundry
Ease of Install ★★★★★ ★★★★★ ★★★★★
UI Quality ★★☆☆☆ ★★★★★ ★☆☆☆☆
Speed ★★★☆☆ ★★★★★ ★★★★★
Model Variety ★★★★☆ ★★★★★ ★★☆☆☆
Developer Tools ★★★★★ ★★★☆☆ ★★★★★
Best Use Case Agents, scripting Daily chat, general use High-speed dev + NPU


Conclusion

If you're exploring local AI, any of these tools will work, but they each shine in different ways:

  • Ollama → Best for developers who want a simple, script-friendly tool

  • LM Studio → Best for everyday use with a polished interface

  • Foundry → Best performance on ARM/NPU and ideal for automation

Using the Snapdragon X Elite with 32 GB RAM, all three worked extremely well, with LM Studio and Foundry delivering the fastest response times.


Comments

Post a Comment

Popular posts from this blog

Yes, Blazor Server can scale!

Image
From the very beginning of the release of Blazor Server naysayers have incorrectly stated that Blazor Server can not scale.  They usually state that the limit is 5,000 concurrent users. There are several misconceptions and wrong assumptions about their statement.  Let's break it down and get a better understanding of Blazor server-side scaling. What is the "limit" for users When Microsoft talks about the need to scale an application, what they are saying is that when a normal response for any concurrent user is taking longer than 200 ms, the system is not performant and needs to scale. We will accept that definition: If any normal user response is taking longer than 200 ms, the system needs to scale for concurrent users. By a normal response, we are referring to a page get, a post, etc.  Large database access requests would not be included in this measurement. What is scaling There are 2 common ways to scale, scale, out horizontally, and scale-up, up vertically. In our di...
Read more

Offline-First Strategy with Blazor PWAs: A Complete Guide 🚀

Image
One of the most powerful features of Progressive Web Apps (PWAs) is their ability to function seamlessly even when offline. For users, this means uninterrupted access to critical features regardless of network conditions. For developers, adopting an offline-first strategy ensures a robust, user-friendly experience while aligning with modern web development best practices. Blazor, Microsoft’s modern web framework, simplifies building PWAs with offline capabilities by leveraging features like service workers, caching, and IndexedDB. In this guide, we’ll dive into how you can implement an offline-first strategy in your Blazor PWA. Why Offline-First Matters An offline-first strategy doesn’t just improve user experience—it also provides: Reliability : Apps remain functional even with intermittent or no connectivity. Performance : Cached assets load faster, improving responsiveness. Engagement : Users can interact with the app at any time, enhancing retention and trust. Key Components of Of...
Read more

Customizing PWA Manifest and Icons for a Polished User Experience 🚀

Image
  Progressive Web Apps (PWAs) blur the line between web and native applications, offering a seamless and engaging user experience. But to make your PWA truly stand out, it’s essential to customize its manifest file and design professional-grade icons and splash screens . These elements influence how users perceive your app and ensure it looks and feels like a polished, high-quality product. In this guide, we’ll explore how to tailor the manifest.json file and create stunning icons and splash screens to deliver a professional and consistent user experience. What is the PWA Manifest File? The manifest.json file is a JSON file that provides metadata about your PWA. It defines how your app appears to users and controls its behavior when installed on a device. Key properties include: Name and description : How your app is displayed to users. Icons : App icons in various sizes. Theme and background colors : Define the app's look and feel. Display mode : Controls whether the app appear...
Read more