A week with the RTX Spark: Everything we learned using Nvidia's new laptops

Nvidia's RTX Spark N1X processor is finally here, and we've collectively tested the powerful laptops that feature this new "superchip." RTX Spark is designed to be a high-performance Arm-based SoC for small laptops and desktops that competes with the best from Apple, Intel, AMD and Qualcomm. It has many cores, Nvidia […]

Nvidia's RTX Spark N1X processor is finally here, and we're collectively testing low-end power laptops with this new "superchip." RTX Spark is designed to be a high-performance Arm-based SoC for small laptops and desktops that competes with the best from Apple, Intel, AMD and Qualcomm. It has plenty of cores, Nvidia RTX Blackwell graphics, and enough power and integrated memory to run impressive local AI models.

The flagship is inside Microsoft's Surface Laptop Ultra with 128GB of integrated storage, which our own Tom Warren has thoroughly reviewed. Sean Hollister tested the 64GB version of the Asus ProArt P16, which I also tested in its high-spec 128GB configuration. Because I'm a badass, I have a 64GB Lenovo Yoga 9n 2-in-1 and a 32GB HP OmniBook Ultra for testing.

All of these laptops feature an N1X chip with 20 processor cores and 6,144 CUDA GPU cores, and they're all very expensive—from the HP OmniBook at $3,199.99 to the Asus ProArt at $6,999.99. There's an 18-core/5120 CUDA N1X chip in cheaper configurations, but it's only $2,599.99 cheaper than the smaller Surface Laptop Ultra or Asus ProArt P14 with 24GB of RAM.

These are large, expensive laptops built on new chips. They are intended for developers, programmers and content creators who may want the ability to use local AI and even play games using the latest Nvidia technology such as DLSS 5. There's a lot to test, and we haven't had it yet. Time to test it all. So below you won't find a comprehensive rated review of a single laptop, but instead our impressions and thoughts from three writers who used five laptops for a week.

Is Nvidia's "superchip" really great? Have Apple, Intel, AMD and Qualcomm officially announced? Have any of us slept?

[Image: A $7,000 laptop works better and works well.

Antonio Di Benedetto: Windows on Arm is no longer a completely poor experience like it was a few years ago. Thanks in large part to the push on Qualcomm's Snapdragon X chips, there's been a lot of development in native Arm-based apps for Windows. And many x86 programs work well through Microsoft's Prism emulation layer.

Adobe has had the original versions of Photoshop and Premiere for a while now, and even though my treasured classic Lightroom is still an x86-based program, I don't see any issues using it on the Arm. At this point, you'll either have to rely on specialized x86-only Windows on Arm software to be a non-starter - or you'll want to freely play any game without worrying about incompatibilities.

Sean Hollister: Every once in a while, there's a time when an app takes longer to load than it should, and I don't know if it's going to load at all. Even then, it usually loads. Sometimes, it takes more than one test. Evernote notes disappeared a few times a day when I alt-tabbed, but a force refresh fixed that. And apps don't seem to respect my system's preference for dark mode.

But I largely agree with Antonio. I've made an RTX Spark my daily driver with all my usual work and personal apps—Signal, Discord, Audacity, CapCut and Evernote, Irfanview, Slack, ScreenToGif, and more—and nothing has ever let me down. I never had to think about whether I was running an Arm or x86 program. As you will read below, there were real problems with the game and the artificial intelligence.

Tom Warren: I've been using Windows daily on an Arm laptop with a Qualcomm chip for the past two years, and things have improved to the point where I can no longer tell if a program is native or not. It's the same story with the RTX Spark laptops, and I feel Nvidia's entry into the category only improves the story for creative applications and gaming.

[Image: Gears of War: E-Day running on a Surface Laptop Ultra during Microsoft's demo.

TW: Windows on Arm has never felt like a serious gaming platform before. While there are still no native Arm games, Nvidia's GPU helps bruteforce through the emulation layer. Your in-game mileage will vary depending on the title, but for the most part I've found that medium settings, running at 1440p and DLSS performance will get you around 60fps in most games.

Most of my gaming is on my desktop, where I'd crash every day using the RTX 5090 on a 32" 4K screen at max gaming settings, but I didn't mind the look at medium settings on a 15" screen. I was able to run Black Myth: Wukong at 70fps at native resolution (3270 x 2180) with medium settings and DLSS performance mode enabled. I ran the same medium settings and DLSS performance in Cyberpunk 2077 and averaged 66 fps, and Forza Horizon 6 managed an average of 57 fps.

In some titles you may need to lower the resolution or settings. It's definitely not as good as a gaming laptop, but for something on the same level as a MacBook Pro, it's good enough for gaming on the go.

SH: As Tom says, the GPU power is there - but the CPU might hold back some of your games. I had plenty of power for a smooth 60fps on Returnal (at 1440p on high settings), Expedition 33 (on DLSS "Performance" enhanced 1200p on high settings), Jedi Survivor (DLSS "Performance" enhanced 1200p on medium settings) and Rady180p on the highest internal settings).

But when I ran Helldivers 2, even downgrading to a 1080p DLSS didn't guarantee 60fps on the highest difficulty missions. There are too many enemies on the screen for Spark to keep up. Star Wars Zero is even more CPU-bound — the Spark didn't care if I maxed or minimized the graphics and screen resolution. No matter what I did, I usually limited my framerate in battle to between 37 and 39 fps.

Even ancient and low-powered games can have weird features. I tried the 2003 title Max Payne 2 and saw conditions where the frame rate dropped to 50 fps. I even saw some problems with the excellent UFO 50, a collection of 2D games, although a few other 2D games ran perfectly.

I'm also not sure if the RTX Spark is suitable for VR. Titanfall 2 VR's great mode worked and seemed to work well at first, but after a few minutes it lost a bit of stability. It could be my Wi-Fi, as the Steam Frame Valve wireless adapter driver won't install on the Spark. Neither did Steam's streaming audio solution, so the audio came out of the Nvidia laptop instead of the headset.

Just to be perfectly clear, this is all assuming you want a laptop that maxes out its fans and dissipates heat positively onto your desk or feet. That's not unusual for thin gaming notebooks, but it's louder and hotter than a good desktop or handheld.

AD: I've wanted a MacBook Pro-like laptop with full PC gaming parts for as long as I can remember, so I was excited for these laptops to deliver the goods graphically. The first night, I fired up the Asus ProArt P16, with all its 128GB of super-large memory, and when I fired up the Big Walk for a planned online night with some friends, my first thought was, "Well, it looks good. I guess." I know Big Walk is not a triple graphics titan, but it has some pretty tough graphics. The "slope buddy" genre. I expected the N1X to max it out, or at least come close, but not really.

I had to downgrade the P16's 4K display to 2560 x 1600, and instead of the higher or very high Big Walk graphics settings, I turned it down to High to keep a nice steady 60fps. Does any of that really matter in this silly co-op game about solving puzzles with your friends? Of course not. It looked good after all. But it was a sobering reminder that this "superchip" is no graphics wizard.

As Tom and Sean point out, the RTX Spark can play decent games, but in some cases lower your expectations to medium or even low settings and be prepared to rely on DLSS and frame generation. There's no denying that cheaper desktops and even laptops with Nvidia discrete graphics can easily beat the N1X. Nvidia claims it's like having an RTX 5070 laptop GPU in the SoC, but in my testing it was somewhere between that and an RTX 5060 laptop GPU – certainly not a gamer, but if you're a gamer first, your money will go much further with traditional Nvidia graphics cards.

[Image: Unfortunately, only Microsoft's Surface Laptop Ultra features USB-C soft magnetic charging.

SH: Nvidia's partners saw fit to equip their first wave of RTX Spark laptops with some of the biggest batteries you can bring on a plane without question - but even with a 99.9Wh pack like that. In my Asus ProArt P16, these laptops won't win awards for longevity. While Tom and Antonio spent their days testing every use case under the sun, I made the P16 my daily driver for routine tasks while watching the battery meter like a hawk. It's mostly about writing stories in a web browser, and it doesn't quite last that day.

My best run was 5 hours and 40 minutes of real-world work at 50% brightness, including an hour-long video call with Tom and Antonio to strategize our RTX Spark overlay program. The next day, I only got about 5 hours out of a full charge because I spent two of them editing this video in Capcut. The third time I drained the battery, it took 5 hours.

With web work, I usually see 13-17 watts of battery drain at the laptop's default settings, but with light video work it goes up to 20-30 watts, 50 watts while exporting Capcut (short) video, and I got 120 watts playing Indiana Jones and the Great Sear. The battery just couldn't feed at that rate - it went down slowly and at 30% charge I saw great performance as it dropped dramatically to 60W.

The good news is that the Asus charges quickly from the 140W USB-C PD power supply. It took 20 minutes to reach 50% battery charge, another 15 minutes to reach 70%, and a total of 1 hour and 15 minutes to reach 95%.

AD: I had to run a lot of benchmarks while plugged in, so I wasn't able to spend as much time testing the battery life as I usually like. Anecdotally, battery life seems consistent in most cases and starts. As Sean saw, you're unlikely to get through a full eight-hour workday unless you go out of your way to be efficient. And if you shoot something heavy like a local AI model (more on that in a minute), you won't even make it to lunch. Finally, as I continue to use these laptops, I'll do a more thorough battery test, including our usual summary tests for a quantitative 1:1 comparison.

[Image: A local AI model running on a Surface Ultra laptop.

TW: Not yet. All of this currently looks like a beta version aimed at developers: from the experience of installing AI agents that elevate the command line and Visual Studio, to fans looking for the most basic queries.

I tried to find compelling use cases for the power of these chips for local AI tasks. By using a free cloud model, many tasks can be done faster and without spinning laptop fans. The power here is really in letting the agent run at full speed to build something overnight, running tasks that developers love.

It's also nice to have all this on a local device, without having to share emails and files with a cloud service and worry about privacy aspects. However, I regularly close OpenClaw or Hermes and don't let them run in the background all day, especially if I've given them access to local files. I'm just not ready to fully trust AI, local or unlicensed.

Microsoft and Nvidia both envision a future where AI agents automate tasks on your PC, but I'm still not entirely convinced. While there are some everyday tasks that I can certainly automate, I like to use my PC to edit photos or videos, write, and play games. I don't want to sit there and watch a robot do it all for me. That's part of the fun of computing for me.

SH: I came out of Nvidia Editor Day thinking that I might finally have a reason to enable local AI. I asked Hermes, running Qwen3.8-27B, to create a daily news report for me, similar to the one in the Nvidia demo on stage. It took me 10 minutes and 10% of my total 99.9Wh battery to finally compile a list of stories that took cloud-based Opus 5.5 in half - with my GPU completely monopolized and the Spark laptop heating up dramatically. As long as I kept Hermes open, the laptop was hot and a little slow.

Cloud also produced better results with more detail and better resources. I guess I got local AI tokens for "free", I don't trust the cloud with my email, and I could probably use a smaller, lighter model. But I went with what the program recommended and this is the result. I tried reinstalling Hermes to make sure it wasn't misconfigured, but this time, the request took even longer and drained 60% of my battery. oh

AD: I'm still early in my journey to learn how to use local AI, and these RTX Spark laptops are obviously a viable – albeit expensive – option. But even if you have 128 GB of memory to use some bulky models, it's still not a user-friendly experience for beginners. Microsoft's introduction of Get Started in Windows 11 gives you easy one-click installation for Hermes Agent, OpenClaw, and Perplexity. (They all have an ESRB rating for some reason?) But after that, you're kind of on your own. I stuck with Hermes because that's what I used to use on Mac Studio, but if you're using it on a system with 32GB of RAM or less, it'll just tell you there's not enough memory for the local Qwen models it offers - either point you to the cloud or completely rely on you to get a smaller model yourself.

Using local AI has significant privacy advantages over cloud models, but it's certainly more convenient because local LLM processing is much slower and more taxing on your system. Creating a local model will tear through your laptop's memory through the large batteries of any of these laptops, and the results from LLM are still different. I burned 30% of the Lenovo Yoga 9n battery in 36 minutes to get a Qwen 3.8 27B model to resize and add a battery. White background to five product images in PNG format - and it cropped the images incorrectly and cut off the products. I did this immediately after following up a few times and grabbing the power cable to avoid killing the Hermes laptop in the process.