Melvin Ryan  Fetalvero

Review: Asus ExpertBook Ultra – Pushing Local AI Limits 

Article (news)

ABS-CBN News

<p>MANILA — Back in college, when I was assembling my own desktop PC for personal and academic tasks, Intel and Asus were my go-to brands. For a long time, these were the trusted favorites for software engineers, especially here in Asia. I felt a rush of nostalgia and excitement when I finally got my hands on the new Asus ExpertBook Ultra. </p><p>If you are interested in the technical specs of this flagship product, you can check them out here: <a href="https://www.abs-cbn.com/news/technology/2026/7/1/review-asus-expertbook-ultra-is-a-featherweight-laptop-that-packs-ai-heavyweight-power-1604">Asus ExpertBook Ultra Review</a>. </p><p>In this article, I will focus on testing the laptop across various real-world use cases so that interested readers can make an informed decision on whether this machine suits their needs. </p><div><strong>Design and Build</strong></div><div><span class="fr-img-caption fr-fic fr-dib" style="width:960px;display:block" width="960"><span class="fr-img-wrap"><img src="https://od2-image-api.abs-cbn.com/prod/editorImage/17875823438641782873943105asus expertbook ultra 2.jpg" style="width:960px" class="fr-fic fr-dib" width="960"/><span class="fr-inner" style="text-align:left">Asus ExpertBook Ultra. <em>Arthur Fuentes, ABS-CBN News</em></span></span></span>Having used MacBook Pro aluminum unibodies for years, I was skeptical about how another manufacturer would measure up—but the ExpertBook Ultra pleasantly surprised me. Cramming a high-TDP Intel Core Ultra X7 processor and 64GB of RAM into such a slim, travel-ready chassis is a massive thermal balancing act, yet Asus pulled it off. The build quality is rock solid; the deck has zero flex, which is a relief when you’re furiously hammering out code against a deadline. </div><p>What truly impressed me was the efficiency of the design. Typically, adding a touchscreen and pen digitizer makes a laptop bulky, but Asus kept the footprint aggressively thin while effectively venting heat out the back. It’s a thoughtful detail that keeps your wrists cool, even during heavy Docker builds and LLM inference tasks. </p><p>Powered by the new Intel Panther Lake architecture, this machine delivers a staggering 122 TOPS of AI throughput, fitting desktop-class power into a 1.1kg frame that meets military-grade durability standards. After putting this under a sustained load, I’ve had to rethink what &quot;premium&quot; really means—especially when the 14-inch 3K OLED display remains crisp and bright, regardless of how hard I push the system. </p><div><strong>Benchmarking: &quot;Raw&quot; vs. &quot;Real&quot; </strong></div><p>Synthetic Tests: <strong>Geekbench CPU Tests</strong></p><div><span class="fr-img-caption fr-fic fr-dib" style="width:960px;display:block" width="960"><span class="fr-img-wrap"><img src="https://od2-image-api.abs-cbn.com/prod/editorImage/1787582609705geekbenchcpu.PNG" style="width:960px" class="fr-fic fr-dib" width="960" alt="Geekbench CPU Test on Asus Expertbook Ultra"/><span class="fr-inner" style="text-align:left">Geekbench CPU Test on Asus Expertbook Ultra. <em>Melvin Ryan Fetalvero</em></span></span></span></div><div><strong>CPU Comparison: Business &amp; Pro-Consumer Flagships</strong></div><table><tbody><tr><td>Device Category</td><td>Model / Processor</td><td>Single-Core Score</td><td>Multi-Core Score</td><td>Main Advantage</td></tr><tr><td>User Device</td><td>Asus Expertbook (Panther Lake + B390)</td><td>2,480</td><td>16,950</td><td>Unmatched Multi-Core: Best-in-class for heavy VM use and local AI.</td></tr><tr><td>Business Rival</td><td>ThinkPad X1 Carbon (Gen 13/14)</td><td>~2,650</td><td>~13,500</td><td>Enterprise security; cooler/quieter operation but less raw power.</td></tr><tr><td>Power/Creative</td><td>MacBook Pro 14 (M4 Pro - 12 Core)</td><td>~3,400</td><td>~15,200</td><td>Industry-leading single-core burst and extreme power efficiency.</td></tr><tr><td>Efficiency/Thin</td><td>Dell XPS 13 (Core Ultra Series 2 - Lunar Lake)</td><td>~2,750</td><td>~11,200</td><td>Maximum battery life; lower multi-core due to focus on power-sipping.</td></tr><tr><td>AI Specialist</td><td>HP EliteBook Ultra (Snapdragon X Elite)</td><td>~2,800</td><td>~14,400</td><td>Instant-on experience; excellent Windows AI integration (ARM-based).</td></tr><tr><td>Workstation</td><td>ROG Zephyrus G16 (Ryzen AI 9 HX 370)</td><td>~2,850</td><td>~15,100</td><td>Excellent hybrid performance for gaming and creator apps.</td></tr></tbody></table><div><br/></div><p>The standout metric here is the multi-core score of 16,950. This result puts the ExpertBook Ultra well ahead of current benchmarks from the Snapdragon X Elite and Intel’s Lunar Lake flagships in sustained, multi-threaded workloads. It’s a clear indicator that this machine isn&#x27;t just optimized for light productivity; it’s engineered for serious, sustained heavy lifting. </p><p>For the &quot;Professional Power User,&quot; this performance is a genuine game-changer. Whether you’re juggling multiple virtual machines, crunching massive Excel models, or compiling complex codebases while simultaneously jumping on 4K video calls, this machine handles the load without breaking a sweat. It translates those synthetic benchmark numbers into a tangible, seamless experience, ensuring you stay in your flow state regardless of how many background processes are fighting for resources. </p><p>Synthetic Tests: <strong>3D Mark test </strong></p><div><span class="fr-img-caption fr-fic fr-dib" style="width:960px;display:block" width="960"><span class="fr-img-wrap"><img src="https://od2-image-api.abs-cbn.com/prod/editorImage/17875831220003DMark Test Result.png" style="width:960px" class="fr-fic fr-dib" width="960" alt="3D Mark Test. Melvin Ryan Fetalvero"/><span class="fr-inner" style="text-align:left">3D Mark Test. <em>Melvin Ryan Fetalvero</em></span></span></span></div><div><strong>AI Laptop Comparison: Graphics vs. AI Processing Power</strong></div><table><tbody><tr><td>Laptop Model</td><td>Platform (SoC)</td><td>iGPU Architecture</td><td>Steel Nomad Light Score</td><td>Platform AI TOPS (Total)</td><td>Competitive Position</td></tr><tr><td>Asus Expertbook B5/B9</td><td>Intel Core Ultra (Panther Lake)</td><td>Intel Arc B390 (Xe3)</td><td>~5,762</td><td>~120+ TOPS</td><td>Leader in iGPU Power: Offers near-discrete GPU performance in a business thin-and-light.</td></tr><tr><td>ROG Zephyrus G16</td><td>AMD Ryzen AI 9 HX 370</td><td>Radeon 890M (RDNA 3.</td><td>~4,200</td><td>~80 TOPS</td><td>The Versatile Pro: Strong CPU, but graphics fall behind Intel&#x27;s newest Xe3 architecture.</td></tr><tr><td>Microsoft Surface Laptop 7</td><td>Snapdragon X Elite (X1E-80-100)</td><td>Qualcomm Adreno X1</td><td>~2,100</td><td>~45 TOPS</td><td>Efficiency Specialist: Amazing battery, but significantly lower gaming/graphics performance.</td></tr><tr><td>Dell XPS 13</td><td>Intel Core Ultra Series 2 (Lunar Lake)</td><td>Intel Arc 140V (Xe2)</td><td>~4,100</td><td>~115+ TOPS</td><td>Efficiency Focused: Great AI TOPS but fewer &quot;graphics cores&quot; compared to your B390.</td></tr><tr><td>MacBook Pro 14 (M4 Pro)</td><td>Apple M4 Pro</td><td>16-Core GPU</td><td>~6,100</td><td>~38 TOPS (NPU only)</td><td>The Creator Rival: Closest graphical competitor, though total &quot;Platform TOPS&quot; metrics vary by OS.</td></tr></tbody></table><div><br/></div><p>While most current &quot;AI Laptops&quot; are configured with 16GB or 32GB of RAM, the 64GB ceiling in the ExpertBook Ultra is a massive performance multiplier. For the professional user, this isn&#x27;t just about &quot;future-proofing&quot;; it’s about unlocking workflows that are simply impossible on standard configurations. That extra overhead provides the breathing room needed to load complex, high-parameter AI models directly into local memory, drastically reducing latency and dependency on cloud-based processing.  </p><p>This advantage is most evident when running resource-intensive applications on Windows. Whether you are locally hosting large language models (LLMs) that require massive context windows, running complex simulations, or gaming while simultaneously managing heavy development environments, that 64GB footprint keeps your system responsive where others would begin to stutter. In short, it shifts the laptop from a device that just accesses AI tools to a mobile workstation capable of creating and training them. </p><p>Real-World Testing: <strong>Gaming and Creative Workflows </strong></p><div><span class="fr-img-caption fr-fic fr-dib" style="width:960px;display:block" width="960"><span class="fr-img-wrap"><img src="https://od2-image-api.abs-cbn.com/prod/editorImage/1787583451842Adobe Premiere Test.png" style="width:960px" class="fr-fic fr-dib" width="960" alt="Editing Videos and Exporting them to other formats was a breeze with the Intel Arc B390 GPU in this laptop. Melvin Ryan Fetalvero"/><span class="fr-inner" style="text-align:left">Editing Videos and exporting them to other formats was a breeze with the Intel Arc B390 GPU in this laptop. <em>Melvin Ryan Fetalvero</em></span></span></span></div><p>The combination of the Intel Core Ultra (Panther Lake) and Intel Arc B390 (Xe3) punches well above its weight class. In testing, this hardware translated synthetic benchmark wins into tangible performance gains: whether I was gaming with stable frame rates on AAA titles or tackling complex 4K video edits in Adobe Premiere Pro, the system remained remarkably composed.</p><div><span class="fr-img-caption fr-fic fr-dib" style="width:960px;display:block" width="960"><span class="fr-img-wrap"><img src="https://od2-image-api.abs-cbn.com/prod/editorImage/1787583567240Steam Test.png" style="width:960px" class="fr-fic fr-dib" width="960" alt="I tested AAA Games via steams and I was able to game with decent FPS in this laptop. Melvin Ryan Fetalvero"/><span class="fr-inner" style="text-align:left">I tested AAA Games via Steam, and I was able to game with decent FPS in this laptop. <em>Melvin Ryan Fetalvero</em></span></span></span>That stability is anchored by four &quot;Hero Features&quot; that elevate this from a standard business laptop to a mobile workstation: </div><ol><li>64GB of RAM: Provides an enormous buffer, allowing for seamless high-bitrate 4K playback without the need for proxies. </li><li>Xe3 Integrated Graphics: Delivers surprising hardware acceleration for H.265 encoding, keeping export times snappy. </li><li>NPU-backed AI Integration: Streamlines repetitive tasks like auto-reframe directly on-device. </li><li>Thermal Efficiency: The chassis design prevents aggressive throttling, maintaining a consistent performance curve from start to finish.</li></ol><div><strong>Experiment 1: Asus MyExpert </strong></div><div><span class="fr-img-caption fr-fic fr-dib" style="width:960px;display:block" width="960"><span class="fr-img-wrap"><img src="https://od2-image-api.abs-cbn.com/prod/editorImage/1787583740161Asus My Expert Text Analysis and Extraction from Images.png" style="width:960px" class="fr-fic fr-dib" width="960" alt="Using Knowledge Hub app I was able to process Word Documents and Image files for summary and analysis. Melvin Ryan Fetalvero"/><span class="fr-inner" style="text-align:left">Using the Knowledge Hub App, I was able to process Word Documents and Image files for analysis. <em>Melvin Ryan Fetalvero</em></span></span></span></div><p>Right out of the box, this device includes Asus MyExpert software. Its &quot;Meeting AI&quot; and &quot;ExpertMeet&quot; features allow for real-time meeting transcription. The Knowledge Hub allowed me to process Word documents and images, and the experience was reminiscent of Google’s NotebookLM, but entirely local. The performance was solid, though I did encounter some errors when attempting to process very large files, which led me to further testing. </p><div><strong>Experiment 2: Ollama </strong></div><div><span class="fr-img-caption fr-fic fr-dib" style="width:960px;display:block" width="960"><span class="fr-img-wrap"><img src="https://od2-image-api.abs-cbn.com/prod/editorImage/1787584064285Ollama Processing a PDF file.png" style="width:960px" class="fr-fic fr-dib" width="960" alt="Using Ollama enabled me to switch between models in processing PDF files, Images as well as doing some light coding experiments. Melvin Ryan Fetalvero"/><span class="fr-inner" style="text-align:left">Using Ollama enabled me to switch between models in processing PDF files, images, as well as doing some light coding experiments. <em>Melvin Ryan Fetalvero</em></span></span></span></div><p>I installed Ollama and downloaded several local LLMs. Using the Gemma 4: 26B models, I was able to process images to generate SEO and metadata recommendations for articles, as well as summarize PDF documents. For the most part, I achieved approximately 20–30 tokens per second. Larger documents were occasionally problematic due to context window limitations, a challenge that is particularly relevant for journalists and researchers handling extensive archives. It is here where the 64GB of RAM makes the world of difference. I was able to test various models for various use cases. It is important to make sure that Vulkan is configured. </p><div><strong>Experiment 3: OpenCode</strong></div><div><span class="fr-img-caption fr-fic fr-dib" style="width:960px;display:block" width="960"><span class="fr-img-wrap"><img src="https://od2-image-api.abs-cbn.com/prod/editorImage/1787584331765ADB Website Creation Test.png" style="width:960px" class="fr-fic fr-dib" width="960" alt="Using Opencode Agent Harness enabled me to push this laptop further paired with local models in Ollama. Melvin Ryan Fetalvero"/><span class="fr-inner" style="text-align:left">Using OpenCode Agent Harness enabled me to push this laptop further, paired with local models in Ollama. <em>Melvin Ryan Fetalvero</em></span></span></span></div><p>In my previous reviews, Openclaw AI Agentic Harnesses have been widely adopted. For this review, I tested OpenCode, specifically because of the community’s focus on security and safety. Using WSL (Windows Subsystem for Linux) in Windows 11 made the environment familiar and efficient. Utilizing the Qwen3Coder:30B model, I was able to set up a Python environment within Ubuntu. </p><p>OpenCode extended my capabilities with web search, file writing, and shell access, alongside the ability to utilize custom Skills and Agents. This was a game-changer. Using &quot;Plan&quot; and &quot;Build&quot; subagents, I created a Python script to process the PDF files that the previous model could not handle. I then used the extracted text to create summaries and built a simple webpage to host both the summaries and embedded links to the PDF datasets. </p><div><span class="fr-img-caption fr-fic fr-dib" style="width:960px;display:block" width="960"><span class="fr-img-wrap"><img src="https://od2-image-api.abs-cbn.com/prod/editorImage/1787585050523test website 2.png" style="width:960px" class="fr-fic fr-dib" width="960" alt="This is the sample website created from the summaries of the PDF&#x27;s I tested using Qwen 3 Coder 30B Model Running Locally in Ollama. Melvin Ryan Fetalvero"/><span class="fr-inner" style="text-align:left">Using OpenCode Agent Harness enabled me to push this laptop further, paired with local models in Ollama to create this webpage. <em>Melvin Ryan Fetalvero</em></span></span></span></div><p>It took several hours of trial and error to refine the prompts, but I was eventually successful. This experiment proved that with 64GB of RAM, this device is capable of hosting Docker applications for web servers, databases, and language server protocols, making local, offline development and research a reality. </p><p>While this local setup was by no means faster than frontier models running on the cloud (like Claude Code or Cursor), it highlights a critical use case: reliability during internet outages or when token limits are exhausted. Furthermore, power users can combine both worlds—using frontier models for high-level planning and local LLMs for granular, private execution. </p><p>Ultimately, this setup demonstrates that researchers and developers can process their own data within their own infrastructure. This is not just useful for field researchers with limited connectivity; it is a requirement for finance and clinical data research where data privacy and intellectual property are paramount. </p><div><strong>Verdict: Who is this for? </strong></div><p>Ultimately, the ExpertBook Ultra isn&#x27;t just another thin-and-light laptop—it’s a platform for specialized, high-stakes workflows. For <strong>Content Creators</strong>, it serves as a self-contained production studio, balancing the horsepower needed for demanding video editing suites with the local AI capacity to streamline scriptwriting and media planning in one portable package. </p><p>Crucially, the addition of the active pen digitizer and 10-point multi-touch screen is a major workflow accelerant rather than a boardroom gimmick. If your daily grind involves scrubbing through massive, multi-track timelines in Adobe Premiere Pro, making pixel-perfect masking edits in Adobe Photoshop, or drawing up shot lists, the ability to ditch the trackpad and interact with the 3K OLED screen directly transforms a clunky, mobile editing session into a fluid, tactile experience.  </p><p><strong>Software Developers</strong> will find an even stronger ally in this machine. With the overhead provided by 64GB of RAM and the latest Agentic Harness tools, it transforms into a legitimate force multiplier. Whether you are iterating on local models, managing complex containers, or simply offloading bug-fix tasks to local LLMs, the device provides the breathing room to keep your entire development workflow on-device and accessible, even when you&#x27;re offline. </p><p>Finally, for <strong>Data Scientists and Researchers</strong>, the device offers something that cloud-reliant workstations cannot: absolute data sovereignty. With a 2TB SSD and 64GB of RAM, it provides the necessary infrastructure to store and train on large, proprietary datasets entirely on-device. </p><p>This setup allows you to run Jupyter Notebooks and local databases while seamlessly downloading and switching between various AI models for granular data processing. By keeping sensitive clinical, financial, or research data on your own hardware, you eliminate the risks associated with external service providers, effectively bringing the workstation to the data rather than the other way around.</p><div><em>A quick note on power consumption: Local AI processing and massive Docker container compilation are historically battery killers. Because running these high-parameter models locally pushes the CPU, GPU, and RAM to their absolute thermal ceilings, I chose to run my intense benchmarking and agentic OpenCode tests plugged into wall power. This ensured I bypassed any aggressive battery-saving throttles and got an accurate read on the system&#x27;s maximum hardware output. </em></div><div><em>If you plan to execute continuous local inference as I did with Qwen3Coder or Llama 3.3 in the field, expect the Panther Lake chip to shred through the battery quickly. For standard productivity, however, the Intel Core Ultra architecture has historically proven incredibly efficient when those AI constraints are removed.</em></div><div><strong></strong><strong>Related Video:</strong></div><div><span class="fr-video fr-deletable fr-fvc fr-dvb fr-draggable" contenteditable="false" draggable="true"><iframe width="640" height="360" src="https://www.youtube.com/embed/tOQJNQ7fhXk?&amp;wmode=opaque&amp;rel=0" frameBorder="0" allowfullscreen="" class="fr-draggable"></iframe></span><em></em><br/></div>

<p>The Asus ExpertBook Ultra delivers workstation-class power that lets you run demanding AI programs in a thin, light and very tough package.</p>

The Asus ExpertBook Ultra delivers workstation-class power that lets you run demanding AI programs in a thin, light and very tough package.

REVIEW: Asus ExpertBook Ultra pushes the limits of on-device AI

News

Entertainment

Lifestyle

Sports

Metro.Style

More

BINI

Sudoku

Word of the Day

Lotto

Horoscope

Technology

REVIEW: Asus ExpertBook Ultra pushes the limits of on-device AI

Melvin Ryan Fetalvero,

ABS-CBN News

Published August 31, 2026 09:00 AM PHT

|

Updated August 31, 2026 11:17 AM PHT

Read More:

Asus Expertbook Ultra

AI Laptops

Artificial Intelligence

Intel

Consumer Computing

Technology

Local LLM

ABS-CBN Newsletter

Your go-to source for the latest news, entertainment highlights, exclusive interviews, and updates on your favorite Kapamilya stars.

ABS-CBN Newsletter

Your go-to source for the latest news, entertainment highlights, exclusive interviews, and updates on your favorite Kapamilya stars.

Sign Up

I have read and agree to the Terms and Privacy Policy.

Haven't received our newsletter?

Check your inbox to verify and confirm your subscription. If you don’t see it, please check your spam or junk folder.

ABS-CBN is the leading media and entertainment company in the Philippines, offering quality content across TV, radio, digital, and film. Committed to public service and promoting Filipino values, ABS-CBN continues to inspire and connect audiences worldwide.

Corporate

Investors

Governance

International Distribution

Corporate

Governance

Investors

International Distribution

NPC Seal of Registration

Privacy Policy

AI Policy

Terms of Service

Advertise with Us

© 2026 ABS-CBN Corporation. All Rights Reserved.

NPC Seal of Registration

Privacy Policy

Terms of Service

AI Policy

Advertise with Us

© 2026 ABS-CBN Corporation. All Rights Reserved.

NPC Seal of Registration

Privacy Policy

Terms of Service

AI Policy

Advertise with Us

© 2026 ABS-CBN Corporation. All Rights Reserved.