See more imagesRating 4.3 out of 5 stars with 10 reviews
would recommend to a friend
As a professional Developer I think this is an interesting device from my perspective It brings some real GPU power and local AI capabilities to your Windows laptop with a lot of CPU/GPU cores as we have the 20core Grace CPU and the 6144 core Blackwell RTX GPU With it’s 32Gb of unified RAM this is a solid starting point for any development work, however there are limitations on how ambitious you can be with the local AI. Performance This should handle Visual Studio, VSCode, JetBrains IDEs, multiple tabs, Databases, Docker based workflows, and multiple applications with ease.The 20 core CPU can be a great boost to any parallel workflows you are doing such as building large projects, running automated tests, and processing data however its overall performance will depend on how fast the single core is how well the compiler works with it, memory demands, thermal performance and how well the development tools work on this platform.One thing to note with this is that its using an Arm based CPU so wont be the same as using a traditional Intel/AMD Windows machine. Where possible it’s best to try use native ARM64 applications. You can still run x64 applications via emulation however you may run into some issues with native add-ons, older tools, drivers and development dependencies. If you are maintaining cross platform applications it’d be best to run through the whole toolchain before using it as your main development machine. GPU & Local AI With the 6144 core Blackwell RTX GPU this becomes really interesting With the ability to do CUDA, GPU acceleration, AI experimentation and supported ML.Having the CPU and GPU share the same RAM can be beneficial if you want to run language models locally This removes the strict requirement of having separate ram for the system and GPU. However having only 32Gb of RAM will cause some limitations. The OS, IDEs, background tools and AI all require RAM. Large models, longer context lengths and intensive GPU workloads will quickly eat into your RAM. With smaller quantized models, embeddings, experimenting with inference and AI assisted dev tools this will be fine. However if you are wanting to run larger models, intense finetuning or multiple AI tools this would be a good consideration to go for 64 or 128Gb models if you have the budget Just be sure to pick the right amount of RAM initially because its not upgradable. I would also check that your CUDA, framework and libraries work on this ARM64 Windows machine, No point of having a powerful hardware if it doesn’t work well with the current software stack. RAM & Storage With 32Gb of RAM this should be a solid amount for most development work. It should handle your IDEs, local databases, decent container usage, and multiple applications without any lag The RAM is shared between the CPU & GPU With the 1Tb SSD this will work well for your source code, SDKs, dev environment, container images, data sets and general development files. However over time you will find that Docker images, VMs, package caches, Android Emulators and AI model files can take up a surprising amount of space. If you are planning on using this for application development and Local AI experimentation I would keep an eye on your storage usage, You can upgrade it easily. Windows 11 Pro This is a great option for professional development environments this is ideal if you are heavily reliant on Microsoft tools, enterprise logins, Windows based applications and security features. You’ll be able to start off with Visual Studio, VSCode, Git, PowerShell and browser based development. You’ll also want to consider if you will be using WSL, Virtualisation & Containers. However this will depend on how well these tools support Arm64. Not all x86 dependencies will work as expected. Be sure to check whether the tools your using have native Arm64 support. This includes low level tools, database drivers, compiled packages & bespoke build pipelines.Overall this would be a great option for web development, backend services, cloud tools & general application development. If your project requires bespoke drivers for x64 or very specific virtualisation then be sure to check compatibility first. Aesthetics & General Performance The premium nature of this machine makes it a great option for developers that are mostly using it for sitting down at a desk, attending meetings and moving around. The high resolution touchscreen and 3:2 screen is also a big plus for any developers you’ll get a lot of vertical space for your source code, terminal, documentation & debugging It has a very clean look However I would check that it performs well under load. Compiling, GPU intensive tasks and local inference can all be very different to normal office work Check that the fans aren’t too loud, it doesn’t get too hot, battery drains slowly and performs well under load. Pros Powerful Blackwell GPU for AI & accelerated computing 20 Core CPU for potential parallel development Shared RAM for CPU & GPU 32Gb RAM & 1Tb SSD for a great general use development machine. Windows 11 Pro & High resolution 3:2 screen for professional use. Can be used as a general development machine & local AI experimentation. Cons 32Gb RAM may cause issues when running larger AI models, multiple containers, VMs & development tools. Need to consider Arm64 support. Older x64 tools, native dependencies & bespoke drivers may cause issues.GPU performance will depend on how well the software works with it power limits & thermal performance. If you’re doing general web development & cloud hosted services this is a hard sell as an expensive AI focused machine. Overall it’ll depend on how fast your build times are how well your applications work battery life & performance under load. Overall I think this is an interesting machine. The main selling point is the powerful Nvidia GPU, multi-core Arm CPU and shared RAM. This is great for developers wanting to experiment with local AI, GPU development & intensive applications With 32Gb RAM & 1Tb SSD this is a great middle ground for general development, multitasking & smaller local AI projects. However I wouldn’t expect it to have endless RAM for large language models & intensive virtualisation. These will benefit greatly from more RAM. The main thing to consider is whether your apps will work with this. If you can run your IDEs, compilers, containers & AI tools natively then this could be a great machine to use for your daily development. If you are relying on older x64 tools & bespoke drivers then a traditional x86-64 machine may provide a better experience. Overall this is a great forward thinking machine. If you can make use of the GPU & local AI then there is a great reason for this to exist. Otherwise it will depend on how well everything works with your tools.
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall
Denisik91 Posted
✅This is NEW to me! I had experienced diferent many powerful laptops over the years but this feels and means Business. I usually edit all my DJI footage on ROG16 that is great but this Surface Ultra feels different in a way that's hard to explain until you use it. 💻The first thing that stands out is the premium all-metal build. The Nightfall finish looks professional, feels solid, and the touchscreen display is responsive and beautiful. Everything about the hardware feels refined.What really surprised me, however, was the ARM-based architecture and how efficient it feels in daily use. I spent time exploring NVIDIA tools, running local AI models, and experimenting with Bonsai, and it honestly opened a new world for me. This laptop showed me that modern computing is moving beyond traditional gaming performance and into a space where AI, battery life, and efficiency matter just as much. 🚅Performance feels smooth whether the charger is connected or not. That's something I rarely experienced with other laptops. On many high-performance machines, you notice reduced performance once unplugged. Here, I don't find myself worrying about battery percentage or searching for an outlet. 🧪The system stays cool and quiet during normal and heavy workloads. What impresses me most is the lack of struggle. My gaming laptops often feel like they're constantly managing performance, heat, fan noise, and battery life. The Surface Ultra with it's New Nvidia Spark N1X chip feels efficient and confident. 😁 I am currently making the most of this beautiful machine by implementing local AI workflows to assist with my future drone video editing projects. Running these tools locally allows me to keep everything on-device, reduce reliance on costly subscription-based software, and streamline my workflow. The Surface Laptop gives me the flexibility to improve and refine my content directly on the machine while keeping costs under control. 🤔My final take would be that the new architecture prioritizes efficiency alongside performance, creating an experience that feels incredibly fluid and responsive while maintaining excellent battery life. It's one of the few laptops I've used where Power and Efficiency genuinely seem to coexist!
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall
HBCUgrad13 Posted
[PROS] Workload Warrior: The Microsoft Surface Ultra Laptop is awesome in displaying its might. With my local AI agent, I'm able to input requests and get results quickly. At this current juncture, my needs are normally split between generating code for projects and getting assistance with design and editing operations. So far, so good. All-Arounder: AI processing isn't the only trick up its sleeve. This device does well in pure non-AI related tasks. I'm able to import, export, and everything else in between when I'm video editing as if I was on my actual desktop. Sleek Design: The design of this laptop is minimal and smooth. Minimal may seem like its lacking, but it contains the right ports and the right amount of ports too. It feels nice to the touch and is very much a pleasure to use. Finally, it's solidly built. I want to feel like I can take my laptop anywhere without worrying if it'll break easily in a bag, and this chassis is definitely sturdy. Respectable Battery: When keeping things basic and light, the battery holds up well. I'm able to easily get over 8+ hours of use without needing a charge. With heavier programs, it drains quicker, but that's a common issue with a majority of laptops so this observation is negligible. Beautiful Display: Under the hood, the screen is very high quality. It gets super bright and has a sharp resolution. I give it bonus points for supporting Dolby Vision and well as being a touch screen. My only gripe is that it's not an OLED panel. This screen does struggle some with black details and can show some blooming. However, most time it can be difficult to tell unless you're looking for it. [MIXED BAG] Gaming: AI is the main focus of this device, but I really wanted to take a swipe at the gaming performance since it was highly promoted. I was able to play a number of AAA titles like Cyberpunk, but the big issue is compatibility which will take time because this device is the first of its kind. I tried to play Forza Horizon 6 and the game crashed multiple times. To be fair, this was not officially listed as compatible, but I felt like it should've been a title ready at launch given that it's a pillar for Xbox. Another thing that I'm slightly concerned about when observing this laptop for its specs is the 5070 connection. Sure, it has the same amount of CUDA cores as the Desktop 5070, but it doesn't come close in performing like one because of other variables. My experience has been a mixed given that it's more on par with a 5060-5070 laptop range. Games are playable, but there's a number of kinks to work out still. [CONS] CPU Age: The Surface Laptop Ultra is a hefty investment, but its worthwhile when used for the right purpose. With that being said, it performs well but can lag behind the competition in certain respects. Also, gaming performance is compromised because the CPU is tasked with handling the translation of games into ARM architecture so that they can perform. If the CPU was bit more beefier then this would be a perfect laptop that I wouldn't desire to upgrade anytime soon in the next 3-4 years. [CONCLUSION] For productivity and heavy workloads, this laptop is an excellent choice. It works well in various areas of usage, and performs like a desktop at times. For AI purposes and design/editing programs, it's a no brainer recommendation. For gaming, it's serviceable, but still needs work. Overall, I like the product and will be using it for a long time as my daily driver.
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall
HBCUgrad13 Posted
[PROS] Workload Warrior: I had a chance to put the Surface Laptop through its paces, and I was impressed. For my local AI agent, I was able to input requests and get results back quickly, My AI needs are normally split between coding tasks and editing/design assistance. So far, I'm able to get everything done without any hinderance. All Arounder: For pure non-AI related purposes, this device gets the job done. I'm able to import, export, and everything else in between within my workflow as if I was back on my desktop. Granted, my desktop is more powerful, but its hard to argue against portability and flexibility. Beautiful Display: Under the hood, this computer receives very high marks from me. I love watching content on it because the screen gets very bright and the resolution is almost as sharp as 4K. I give it bonus points for supporting Dolby Vision and being a touch screen device as well. Unfortunately, it's not an OLED screen so I have witnessed some issues with dark scenes and blooming. Yet, these can be hard to spot unless you're looking for them. Sleek and Smooth Design: This computer is very minimal. Minimal may seem like its lacking, but this is not the case. It contains the right ports and the right amount of ports as well. Its nice to the touch and very much a pleasure to use. Furthermore, it solidly built. The chassis is so sturdy that I don't worry about having to transport between work locations. Resilient Battery: For basic and simple tasks, this laptop is world class when a charging isn't available. I was able to easily get 9+ hours without needing to plug it into the wall. I would avoid heavier programs as I saw great diminish on the battery, but this is negligible given that its a problem on all laptop hardware. [CONS] CPU Age: The uniqueness of this product comes down to the unified memory, combined CPU and GPU . I'm more concerned about the CPU rather than the GPU. The CPU being used performs well enough, but is 2 years old upon release. Given that some of its direct (and some older) competitors perform better than it in some aspects is a bit concerning as it relates to overall value proposition and longevity. [MIXED BAG] Gaming: AI is the selling point, but I did take a swipe at its gaming capabilities given that it was moderately promoted as well. Playing Cyberpunk was a fun experience, and I didn't suffer any hiccups. However, compatibility is still being worked out for many games. I'm unable to play titles like Forza Horizon 6 which crashed on my multiple times. To be far, it's not officially support yet, but I feel like should've been a launch-ready title since its a pillar for Xbox. Also, marketing drew comparison of the RTX Spark to the 5070, but that can be a bit misleading. Sure, the Spark has the same amount of CUDA cores as the 5070 desktop, but it is very far from performing like one. I have a 5060 laptop and it's more on par with it as as opposed to a desktop GPU due to power and other variables. [CONCLUSION] With the Surface Laptop Ultra, I've found my daily driver. Performance is excellent across the board, but leaves a bit desired when it comes to gaming. Yet, I have confidence that this will improve as the collaborative effort between Microsoft and Nvidia will address first generation woes. I do worry about CPU performance in the long run, but it will still execute as needed for years to come.
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall
KKCNIck Posted
Setup: I have been building out local AI workflows to handle coding, research, and document analysis on my own hardware so I can avoid recurring cloud subscription costs and per token API fees. Because of that, I approached this review by testing the exact daily workloads I normally run on my current setup (a performance laptop with a newer generation Core i7 processor, 16GB of standard RAM, and an Nvidia 5060 GPU with 8GB of DDR7 memory). Getting this review unit updated out of the box took about an hour longer than usual due to initial system and architecture updates. Once running, setting up my local text and agent stack was painless. AnythingLLM installed cleanly, connected to Ollama right away, detected my local models, and launched the agent interface without a hitch. On the flip side, I spent hours trying to get local image generation working with diffusion models and kept hitting roadblocks. Some emerging creative AI tools simply need more time to mature for this new chip architecture so they do not require heavy troubleshooting. Physically, the hardware is thoughtfully designed, especially the compact power brick and magnetic pull away charging cable, which is reversible so you can route the cable cleanly along the left side. Use: The biggest reason to move from a traditional laptop to this unified memory architecture is overcoming the dedicated video memory wall. On my current 16GB RAM and 8GB Nvidia 5060 laptop, running a highly advanced AI model (similar in brainpower to popular online chatbots) maxes the system out at 100 percent RAM and 90 percent GPU utilization, slowing workflows to a crawl. On this laptop, standard single prompt runs use just 50 percent of total RAM while allocating 12GB to the GPU at 100 percent utilization and holding a cool 60 degrees Celsius. However, pushing into heavy multiple document workloads revealed why 48GB or 64GB of RAM would be even better on this machine. Because Windows 11 itself uses about 6GB to 8GB of memory right off the top, sharing the remaining pool from the 32GB total between the system and the GPU starts to feel tight under heavy loads. When I summarized a full 20-page document or compared multiple large files at once, the laptop pulled 12GB for the GPU and drew the rest from system RAM, which noticeably slowed the machine down, spun the fans up loudly, and made the chassis hot. Despite the heat under maximum load, the finish of the metal on this laptop feels premium. It does not feel like hard cold metal, but rather comfortable and sturdy. I also appreciate the fans placed on the undersides of the edges, allowing it to sit on my lap without frying me. The keyboard is phenomenal for typing and is easily one of the best I have ever used. The 15 inch 3:2 display gives me way more vertical space to view code and long AI outputs, and the speakers are easily the best I have heard on any Windows laptop. I love the bright and high-quality display but wish it was OLED and less glossy. For my workflow, I also do not need the touch screen. It would be great to see a gaming centered version in a future iteration or generation, as I appreciate the premium feel but would like better gaming support. Storage performance is one of the strongest aspects of this configuration. While everyday tasks such as launching applications and browsing Windows will not feel twice as fast, creators working with large image libraries, AI models, or high-resolution video assets will benefit from significantly faster file transfers and project loading times. Storage Performance Scorecard (CrystalDiskMark): • Sequential Read: 14.5 GB/s (PCIe Gen5) • Sequential Write: 12.6 GB/s (PCIe Gen5) • Typical Premium Gen4 Drive: Approximately 7.0 GB/s To show how much faster this machine handles daily local AI compared to my current setup, here is how much time it saved across my real-world tests: Local AI Speed Scorecard: • Token Generation Speed: 3x faster on average • Coding a Flight Simulator Game: 3.8x faster • Writing a 1,500 Word Guide: 3.6x faster • Executive Summary Thinking Time: 3.5x faster initial reasoning • Advanced Reasoning Prompt: 3.3x faster • Multiple Step Planning Workflow: 2.8x faster • Huge Output Generation Rate: 2.7x faster output speed • Coding Benchmark Speed: 2.6x faster output speed • Long Context Retrieval: 2.4x faster • Pulling Top 10 Stock Data: 2.1x faster • Knowledge Stress Test: 2.1x faster • Agent Capability Workflow: 1.8x faster Since my goal is replacing paid cloud AI subscriptions with free local models, I also benchmarked this against a fast cloud AI service on real document tasks: Local AI vs. Paid Cloud AI Scorecard: • Math & Short Summaries (10 to 20 pages): 9.9s to 46.2s locally vs 2.0s to 16.8s cloud • Analyzing 150 Pages (Three Guides): 33.0s locally (with better accuracy) vs 10.0s cloud • Full 20 Page Summary & Comparison: 1m 37s to 3m 14s locally vs 5s to 9s cloud While the shared 32GB RAM limit slowed the laptop down on heavy comparisons, the local model delivered a much more thorough analysis without costing a cent in tokens. What really impressed me during those heavy memory stress tests is how fast the cooling system recovers. Even when a big job pushes all resources and gets the fans roaring, the laptop cools down rapidly as soon as the task finishes. The laptop also does an excellent job on standby, and battery life for once does not really affect performance. I like finally using a laptop around the house without plugging it in constantly for peak performance. Based on my drain tests, you can expect excellent endurance depending on your workload: Battery Life: • Heavy Local AI Load (Best Performance): 3.1 hours (32 percent drain in 1 hour) • Mixed Use (Best Performance): 4.1 hours (48 percent drain in 2 hours) • Mixed Use (Balanced Profile): 8.2 hours estimated Pros: • Runs local AI workflows faster • Extremely fast PCIe Gen5 storage • Exceptional speakers and phenomenal keyboard • Comfortable and sturdy metal build • Reversible magnetic USB charging cable Cons: • Shared 32GB RAM feels tight • Display is glossy, not OLED • Needs better dedicated gaming support • Initial setup updates take long • Image generation needs software maturity Summary: If you are looking to replace a traditional laptop to build local AI workflows and cut out monthly cloud fees, this machine delivers a massive speed upgrade and premium PCIe Gen5 storage. While power users running heavy loads would benefit from more RAM, the fast local inference, sturdy metal build, and phenomenal typing experience make it a very capable daily workstation.
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall
NickAtWTF Posted
The Microsoft Surface Laptop Ultra is stunningly capable. I'm about to dive into some super geeky specs below, but here's the short version: this laptop performs as if it has a desktop-class GPU while not tethering you to a desk and is ready for on-device AI-assisted co-programming using some truly hefty models from ollama. Your mileage may vary a bit based on your model of choice, but with mild tinkering and testing, During plugged-in gaming tests, ARC Raiders and Portal with RTX ran smoothly with no stutters at medium and high settings. Cinematic settings in ARC Raiders dropped to around 20-30 FPS. Between the NVIDIA RTX Spark N1X GPU, a snappy 1TB SSD, and some of the fastest memory available to laptops right now, it’s gaming ready in a slim form factor. Its 301 GB/s of memory bandwidth doesn’t match a desktop RTX 3060 or 4070, but it’s incredible for a 15" laptop you can take with you to the park. During my tests, I wanted to test on-board AI co-programming - completing a basic coding test and developing and error-checking a full browser-based game. To do this I ran qwen3-coder:30b, nemotron-3.5-bolt:e4b, and qwen3.8:27b with a 64k context window while paired with Claude Code. I tried and failed to load Nemotron Bolt 74B (around 25GB)... so models around 18-20GB is probably the limit. Upper bounds of tokens per second processed with nemotron was around 150. With qwen3.8 it was around 30... but in the span of about 3 minutes it had rendered a compelling, bug-free browser game with a model that was capable of surgical edits to the code in the same way a cloud model operates (though maybe not as fast). That is: you can run local AI to do tasks you'd normally have to pay a cloud provider for... and you can do it on a laptop. The Surface Ultra is absolutely capable of both loading and quickly parsing qwen3.8 and utilized 17GB of working memory to do it... qwen3-coder apparently also availed itself of 20GB of working memory, going beyond the 16GB I could set in the NVIDIA app while running entirely on the GPU. I also tested out some smaller video editing, audio editing, and graphic design (Illustrator and InDesign with large complex files) tasks and the Surface Laptop Ultra didn't even break a sweat. The stats and features that make the Surface Laptop Ultra your next workhorse: - 32GB of ram at 9,400 MT/s with 16GB of it able to be dedicated specifically to GPU usage via the NVIDIA app (it gives you a slider you can use to set this) - - 1TB SSD - A beautiful 15" 3K mini-LED touchscreen at 2,000 nits HDR (and yes, it's just as accurate as you'd hope it would be - native resolution is 3270x2180). - A full-size trackpad and responsive backlit keyboard - Solid battery life even while running those local AI models - Around 4.5lbs with a handsome nightfall dark gray or silver aluminum body - 3 USB-C ports, 1 USB-A port, an SD card reader slot, an HDMI port, and a 3.5 mm headphone jack - WiFi 7 built-in - 1080p HD face camera for easy log-in and crystal-clear video calls - Decadent audio - Quiet fans even when cooking with AI - A magnetic charger at the plug and the right-side USB-C port so it won't take a dive if you trip on the cord Bottom line: Microsoft has absolutely nailed it with the Surface Laptop Ultra. For programmers, creatives, and gamers looking for a serious upgrade to their daily driver, this is a fantastic option. The fact that you can also get one in a 128GB ram variety opens up some wild possibilities for some of the most complex local models to be run on something you can carry anywhere... but even the modest 32GB version is capable of things that normally would require a desktop.
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall
psyclopps Posted
I think the best way I can describe this laptop is as a jack-of-all-trades, master of none. This isn’t necessarily a bad thing, just a very good laptop for pretty much everything, but outclassed in specific fields by other brands and architectures (this review is not a comparison). It is great for a power user who wants the ability to do pretty much everything with above average results, doesn’t really lag in any specific area, but is not going to be class leading in any field. The only reason I am giving it three stars is because of the numerous small issues listed in the review and no class leading marks in any area (aside from audio and input). It is a laptop I love very much but it has various issues to be aware of. All tests are plugged in, high performance unless noted in the test that it is on battery. Unboxing: This laptop is PREMIUM. The body feels amazing. First use of the keyboard and trackpad are perfect. Setup is your usual Windows setup with some updates likely needed the first time. For some weird reason, mine took an hour and a half to fully update which was weird since this is much slower than any other Windows laptop I have ever used during setup. Once it is set up though, the first impression is that this is an impressive device. Make sure to fully update all apps and Windows Update items after setting up if anything is leftover. Specifically, update the Nvidia App as this is needed to get full control of your unified memory to the GPU if needed. I am a developer, amateur photographer and occasional gamer so these are the test cases I used this laptop with. RAM: As mentioned in the intro, this laptop gets pretty good results in everything but does not excel in any specific area. My unit is the 32GB model which somewhat limits what I could do with some test instances such as AI. Of note, it is worth knowing that the 32GB is partitioned in various ways to allow it to be shared with the GPU and system. For instance, I set up 8GB to be dedicated to VRAM. Around 1.1GB appears to be reserved by the system for what I assume is NPU/UEFI/Firmware/WDDM and various other items. So, in my instance, the breakdown is: 32GB – 8GB VRAM – 1.1GB System reserved = 22.9GB Usable leftover You can adjust the dedicated VRAM in the Nvidia App but keep in mind, the reserved amount is a bare minimum dedicated to the GPU, not an upper limit as it is allowed to spill into remaining RAM if needed. For most people I recommend just putting it at the bare minimum of 4GB and let the system manage it from there. If you are a creator using software for video editing or 3D graphics, set it higher between 12-16 (16 is the max on this 32GB unit) to ensure resources are always available. I am not complaining as I love this laptop, but if I were to purchase this laptop, I would get the 48GB model minimum to allow 32GB for system use and 16GB to be always set for the video card. I say this as this unit was sent to me as a review unit so 32GB is what I received for these tests. AI: 32GB for LLMs is somewhat restrictive but still doable. 32GB is wonderful for most tasks but AI is somewhat limited at that size. I tried multiple models in LM Studio, but the best mix of accuracy and speed I could find which fit into available RAM space was qwen3-coder-30b Q3_K_L which weighs in at around 14GB and consumes around 17GB of RAM when in use depending on your settings and context size. I chose this model as my goal was to keep roughly 16GB available for the system so I can run instances of VS Code, Docker, PostgreSQL and so on. You could fit a larger model into the 32GB but beware as you will be limited in other tasks if you do. I was impressed by the Tokens per second I was able to get on this model for a Windows laptop. After some configuration I was able to achieve the following: Qwen 3-coder-30b 84.87 Tok/sec Q4 KV cache, 79.91 on battery balanced mode. Qwen 3-coder-30b 85.35 Tok/sec Q8 KV cache, 79.07 on battery balanced mode. When using this model, the fans will spin up for a lengthy response, but it quickly cools down almost immediately after the prompt finishes. These numbers are very good for “integrated” graphics, but you may be better off getting a PC with a discrete GPU and 16GB of VRAM as it will give you better numbers if that is your priority while not touching your system RAM. If staying cool, battery life and maintaining most of your performance while on battery (coffee shop work) are important to you while using an LLM, RTX Spark is a great option. Display: This is going to be the weak point for this device. It is by no means a bad display, it just happens to be the most reflective glass I have ever seen! There is basically no way to use it in dark mode without lots of reflections if there are any other lights in your room. Using light mode is a must in most environments. Otherwise, the display uses Mini-LED which I have few if any concerns about. It looks great, is extremely sharp and bright and accurate for colors. For photography, the colors are spot on and I feel that any designer or artist would enjoy it. Obviously, the drawback of Mini-LED is that blacks are not fully black so that may be an issue for some, but it appears to me the reflections are a bigger issue compared to the black levels. I am unsure if it is all units or just mine, but I can see a slight blue tinge on the right and left edges when displaying bright content which may be an issue for some. Additionally, a drawback for some may be the pixel response time. According to web resources, the response time black-to-white is measured at 57 ms. Most people likely will not notice, but this is poor for anyone who needs fast response times such as gamers. It can be seen when moving the mouse quickly in some instances. The 3:2 aspect ratio at a high point with a high resolution of 3270x2180 with HDR support. I found it outstanding for coding with those specs as you can view differentials easily with extra rows due to the taller aspect ratio. Battery and cooling: The battery life of this device is above average, but as with other areas, not groundbreaking. If you are coding all day, working on office documents, doing edits to photos, it will easily last all day running cool and quiet. I tend to keep it in Balanced battery mode for a good mix of performance which barely differs from High Performance and battery life. If you are a user who is hoping to use LLMs, do 3D design and rendering while on the go with battery, you technically can, but your battery life will be far shorter lasting only a few short hours as the fans will be running with these heavy workloads. The GPU is integrated and almost as powerful as a RTX 5070, so it is more efficient, but also still a relatively high powered part when used to its fullest. Many times I found myself using the laptop running my normal work and later realizing it was on battery since I had zero noticeable difference in performance. Audio: This laptop has probably the best speakers on a Windows laptop currently. It is great if you want to listen to music while working, loud booms from games or just want to hang out with some tunes in the background. Keyboard/mouse: This laptop also has some of the best input experiences out there. The keyboard has plenty of travel and great spacing allowing for outstanding accuracy when typing. The FN key can be toggled on, and off which is always nice. The haptic mouse is extremely accurate and has great response with clicks. If you are working away from your desk, this is the laptop to do it on if you do not have a mouse available. The keyboard is a fingerprint magnet though, especially the spacebar. ARM optimization: For my use, I found almost no issues with compatibility as Windows on ARM has come a long way. The only issues I had were a few games and using SQL. Of note, some software requires updates or beta software to use properly (e.g. Blender 5.3 Beta) which should have full support for RTX Spark. In task manager details, sorting processes by architecture shows that almost every app and service I used are optimized for ARM natively with only a few running x86 emulation (screenshot attached showing all the x86 items which run in my normal workflow). Here is the suite I tested for my workflow: Python, Notepad++, VS Code, Anti-Gravity IDE, LM Studio, Adobe Lightroom, Docker, various Python libraries, Oracle VirtualBox. None had any issues running. Some of the games I tested had some stuttering due to the Prism emulation, but most newer games ran fine. Repairability: This laptop is a dream for anyone who wants to do repairs if needed. Not much is upgradable, only the SSD which is a standard 2280 size. It is a PCI-e 5 hard drive so make sure to use that when upgrading to maintain the fast SSD speeds. Getting to the internals simply involves prying off the three rubber feet which use zero adhesive (thank goodness) and go back on easily. After that you need to remove four T5 screws. Use a suction cup to pull on the back cover and it comes off extremely easily as it is held on with magnets, no clasps to pry off! Once you are inside the system, every single component tells you what size screws are used on it with little labels. Benchmarks: I ran a suite of various benchmarks to test performance of graphics and the CPU. - CrystalDiskMark Read: MB/s 14478.67 MB/s & Write: 12674.09 MB/s - Doom Eternal: RT on, Ultra Nightmare 1440 resolution 100FPS avg - Geekbench AI: Limited options here only supporting Onyx on CPU Single 1574/ Half 3149/ Quantized 6032 - Geekbench 7 GPU CUDA: 184026 (No fans) Battery balanced: 173699 - Geekbench 7 CPU 2528 Single / 22485 Multi - Lightroom export of 156 24MP RAW photos to JPEG 01:50:41
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall
Xephyroth Posted
📦 UNBOXING: Inside the box, you’ll find the Surface Laptop Ultra, a relatively compact 140W USB-C adapter, and the magnetic USB-C charging cable. The charging cable is a bit shorter than I’d probably prefer, but not a big deal. 💪 BUILD: The build quality is notably premium with its aluminum body and enclosure, and weighs in at 4.41lbs. So while it’s not exactly lightweight, it’s also not going to be as heavy as some gaming laptops. One unusual design detail is that the lid lacks a lip or notch, making it a bit trickier to lift open. Notably, it also features user-serviceable parts with online service guides with instructions and pictures on how to conduct disassembly and reassembly. 🔌PORTS: On the left, you have an HDMI 2.1b port, 2x USB4 (40Gbps) Type-C ports, and a 3.5mm audio combo jack. On the right is the USB4 (40Gbps) Type-C port with Magnetic Connect, a USB 3.1 (10Gbps) Type-A port, and an SD Card slot (UHS-I). Overall, the port selection is some of the best that Microsoft has offered yet. But for those looking for the best I/O, the fact that it doesn’t support Thunderbolt 5 and the SD Card slot doesn’t use SD Express can definitely make it feel pretty last-gen for the price. 🔧 EXPANSION & UPGRADES: For upgradability, it’s important to note that the only upgradeable component for the Surface Laptop Ultra is the single m.2 2280 PCIe 5.0 NVMe SSD slot. While it comes pre-installed with a fast 1TB Micron SSD (MTFDLBA1T0THJ-1BP1KABMB), if you decide to upgrade to a larger drive down the road, you can do so. Despite the lack of upgradeability, the Surface Laptop Ultra does otherwise feature a lot of user-serviceable components should something go awry. However, only time will tell if Microsoft is going to have enough inventory of spare parts for customers to purchase. ⌨️ KEYBOARD & TOUCHPAD: The keyboard has been great for me. In a monkeytype test, I was able to achieve 104wpm at 100% accuracy which is a great result for me, overall. I never felt like I needed to acclimate to the keyboard. The keys are backlit and have 100 stops of brightness if you use your mouse cursor or touch screen to adjust the brightness. Otherwise the keyboard macro (on F5) has 3 stops of brightness or can be turned off entirely. Microsoft used a precision haptic touchpad for the Surface Laptop Ultra, and it’s been great to use. If you’ve never used a haptic touchpad before, they are much more consistently responsive compared to mechanical offerings. It’s also quite large so I never feel like I have to swipe over and over just to move my cursor across the screen when navigating across windows, documents, code, or my browser. 📺 DISPLAY: The 15” Mini-LED touch-screen Pixelsense Ultra display features a taller 3:2 aspect ratio with a resolution of 3270x2180 and maximum refresh rate of 120Hz while also supporting dynamic refresh rate between 60Hz and 120Hz Despite its 2,000 peak nits brightness in HDR, it only has 1,344 dimming zones which is significantly lower than competing Mini-LED laptop displays. While in SDR, it can reach an impressive 1,000 nits brightness. However, local dimming does not reliably engage in either SDR or HDR, which completely undermines the purpose of a Mini-LED display. As a result, it can often make the experience of watching content feel like you’re looking at a cheap edge-lit display. Response times and ghosting were also surprisingly some of the worst I’ve seen in a while, especially on a laptop this expensive. If Microsoft couldn’t be bothered to implement local dimming for their display, I’m not sure how much I’d trust that they would try to improve the response times. It makes text blurry in motion which can be distracting when scrolling documents, playing games, or watching videos. 📹 WEBCAM & MICROPHONE: The webcam uses a 2.1MP sensor so you can expect 1080p image quality. Overall, it’s not very sharp and you’d expect something at least 1440p for this price. In this day of online video meetings, having a good webcam can be very important for some people. Microsoft claims their camera supports Windows Studio Effects for portrait blur and auto framing, but it was not available on my unit, and I have installed every update that was made available to me. You can use Nvidia Broadcast which is certainly a superior alternative anyway at the expense of some GPU power, but I found it odd that Windows Studio Effects was completely missing. The built-in dual studio mics sound clear and should be great for meetings and recordings, and the Windows Voice Clarity feature can help isolate your voice from background noise reasonably well. The microphone records 2-channel 24-bit @ 48kHz with no other options to select from. 🔊 SPEAKERS & HEADPHONE: Microsoft’s Quad Omnisonic speakers with Dolby Atmos sound great on the Surface Laptop Ultra. While I don’t think they’re quite as impressive as the speakers in the ProArt P16 or a MacBook Pro, they do get plenty loud, have a wider soundstage, and deliver some actual bass and clarity. After downloading the Dolby Access app, you can enable or disable Dolby Atmos for the speakers and change the EQ as you like. Unlike a lot of other speakers, disabling Dolby Atmos does still sound good, but you can also use an EQ in the Dolby Access app. This applies for the headphone output, as well. The headphone jack uses the low power Realtek ALC721 which should be fine for some basic low-impedance, high sensitivity headphones or earbuds, but it won’t be great for driving some power-hungry planar magnetic headphones. It’s a bit disappointing for a laptop of this price, as much cheaper laptops have better headphone jacks. On another note, none of my headphones with support for Bluetooth LE Audio worked correctly on the Surface Laptop Ultra. Only the left channel of audio would play in both ears, leaving me without stereo sound whenever the microphone was engaged. ⚡ PERFORMANCE: This model of the Nvidia RTX Spark N1X has a 20-core CPU and a 6144-core GPU. The 20-core ARM-based CPU is unfortunately based on a 2-year old design, so its raw single-core performance doesn’t quite compare to some newer ARM-based and x86-based CPUs, but it’s also not so far off that I would find it to be a hindrance for my needs as a software developer. Nvidia claims that the GPU should yield RTX 5070 level of performance, but that isn’t quite true. The N1X has a very different power budget for its GPU than either a laptop or desktop RTX 5070. For gaming, the N1X can get close to some lower-power laptop RTX 5070’s, but because the CPU’s single-core performance is a bit weaker and most games will run under the prism translation layer to run on ARM, there is a performance loss especially in 1% lows. While this isn’t a gaming laptop, you can certainly do some gaming. Where the N1X should excel the most is for local AI. While the N1X’s LPDDR5 memory can’t compete with the GDDR7’s memory bandwidth, its unified memory architecture allows the GPU to access a much larger pool of memory. Unfortunately, the reality is that 32GB of total memory is simply not enough for a good local AI experience on Windows 11 on ARM. With Windows using a large chunk of that memory pool, it doesn’t leave enough left to comfortably run a browser, IDE, and local model. Out of the box, the GPU is configured with 12GB of dedicated VRAM leaving you with only 20GB of system memory for running apps. You can shrink the dedicated VRAM to 4GB or grow it to 16GB depending on your needs, but it will require a system restart. For meaningful AI models, you would want much more VRAM, and while some apps like Hermes Agent, LM Studio, and Unsloth Desktop detect the unified memory and allow the system to use remaining system memory, trying to run other apps will significantly shrink available memory causing your model’s context window to be depleted very quickly. The PCIe 5.0 NVMe SSD, while very fast, cannot make up for the performance loss when 3-bit and 4-bit quantized models have to use swap because there isn’t enough memory left. Multiple times have I encountered graphical glitches because both Windows and my local AI model would be competing for system RAM. So for those who plan to use larger general purpose models, it makes a lot more sense to purchase a model with much more RAM since you’re required to allocate a minimum of 4GB towards the GPU at all times. The onboard Wi-Fi 7 was also very fast on my network, making downloads and network transfers a breeze. However, it should be noted that many external devices may not have drivers for Windows on ARM, and x86 drivers are not emulated unlike x86 applications. Manufacturers must release drivers for Windows on ARM for those devices to work at all, so you may find that some hardware like printers, audio interfaces, or even Microsoft’s own Xbox Wireless USB Adapter does not work for Windows on ARM. 🔋 BATTERY: Microsoft claims 15 hours of local video playback and 12 hours of web browsing with auto-brightness disabled and with the screen at a very dim 150 nits. If you’re a developer and running a lot of power-demanding workloads, you can expect much less battery than this, but I think general mixed use should be able to get probably close to 9-10 hours as long as you’re not cranking the brightness or keeping the display locked at 120Hz without dynamic refresh rate. 🧐 CONCLUSION: The Surface Laptop Ultra doesn’t quite deliver on the “Ultra” moniker with a meager 32GB of RAM that still has to be split between the CPU and GPU through a slider. Not to mention a display that doesn’t perform the way it should. If you’re really committed to the idea of running local AI, you should pursue a model with significantly more RAM, but for the price being paid here, it really feels like an unpolished experience.
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall
Closingracer Posted
This is Microsoft newest attempt at using an ARM SOC with windows and by far the most powerful attempt yet. The GPU is the major focus here with arguably the best gpu for a laptop with this kind of processor. This is the higher end version with a 20 core cpu and the 6144 cuda cores. Specs: Nividia Spark SOC 6144 cuda cores 20 core CPU 1TB SSD 32GB of DDR5 Memory 15 inch 3:2 mini led screen Touch screen Not pen enabled tho Nividia has promoted this device for AI users and productivity the most with gaming on side since this GPU has the same core count as a Nividia 5070 desktop with the unified memory of a max of 128Gb this is a monster potentially for the local AI workflows/creative workflows. I’ve used this computer to do some AI workflows using Perplexity and it has worked great for on device vs using the cloud. I have also used this with adobe premier as well to edit some videos and the experience is getting to the point I would say if you need a portable machine to edit videos on the go on battery it’s up there with the competition almost. I feel like a generation or more you’ll hit parity so you can choose a windows device if that’s what you want.
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall
yogabear Posted
When I say solid I mean solid. At only a little over 4 pounds this feels like something that will last. There is absolutely zero flex with the screen open when you hold it by the very corner. The charger is improved too. It is USBC with a magnetic area that keeps things secure. The display is bright and rich with a very sensitive touch screen. Swipe from the right or left to get access to quick information or commands. Also, it is nice that the internal speakers sound fantastic. According to Microsoft's official tech specs, all three of its USB-C/USB4 ports feature DisplayPort 2.1. Native Support: You can connect up to three 4K external monitors at 60 Hz simultaneously while keeping the laptop's built-in screen active. This came with about a Terabyte of storage, 12 GB of graphics memory and 32 GB of RAM. The NVIDIA RTX Spark processor brings a new level of speed and some refer to it as a "Super Chip." This is truly a desktop computer in laptop clothes! While I typically rely on my dedicated desktop gaming rig for heavy video editing, I transitioned a large-scale project to the Surface. I have no regrets. It matches my desktop's robust performance step-for-step, effortlessly handling a multi-window workflow with countless active elements for seamless multitasking. Complex transition previews render with crystal-clear sharpness and zero perceptible lag, while overall render speeds easily rival the best high-performance workstations I have used. With all this power, I was concerned about cooling. The new thermal system is designed for sustained heavy workloads. This is also built to be serviced. If my experience with Surface devices holds, service will not likely be needed. However, Microsoft informs us that it has removable SSD with replacement parts available through the Microsoft store.
This review is from Microsoft - Surface Laptop Ultra - 15" 3K Touchscreen - NVIDIA RTX Spark N1X 20-Core (2026) - 32GB Memory - 1TB Storage - Nightfall