One big thing changed at Computex this past week: Windows on Arm processors became the next big thing for Windows desktops, specifically for mini PCs designed for productivity and AI applications. So which Arm processor is the best bet?
Until they hit our test bench, we can’t say for sure. But just like your favorite NFL analyst can predict winners and losers during the preseason, we can look at what these chips have already accomplished and start making guesses. The contenders: Qualcomm’s Snapdragon X2 Elite Extreme, which debuted in an Asus mini PC at Computex, as well as a host of mini PCs (including the Surface RTX Spark Dev Box) powered by the Nvidia RTX Spark, the powerful new kid on the block.
Microsoft’s formal support of OpenClaw this past week means that these mini PCs will probably be asked to perform agentic tasks, create AI content, and run productivity applications. Since we know how well AMD’s Ryzen and Intel’s Core processors run in mini PCs, we’re going to largely pass them over in favor of the Arm options. On paper, my guess is that the RTX Spark will be the better machine for AI content creation and gaming, but Qualcomm shouldn’t be counted out; the Snapdragon X2 Elite should have a decided advantage in general productivity and possibly agentic AI. Here’s why.
MiFCOM
CPU performance: Qualcomm likely wins
We know as much about the Nvidia N1X / RTX Spark platform as any other chip. The high-end N1X CPU (made by Mediatek in partnership with Nvidia) consists of 10 Cortex-X925 “extreme” cores, as well as 10 Cortex-A725 “performance” cores, in what is otherwise an off-the-shelf implementation of an Arm processor. Naturally, the selling point of the N1X/RTX Spark is its powerful integrated GPU, with 6,144 “Blackwell RTX” cores, about the equivalent of an RTX 5070 CPU.
Qualcomm, meanwhile, designed its own custom Oryon Gen 3 CPU cores using an Arm architectural license that allowed it the freedom to create its own entirely new design. (It was this license that Arm and Qualcomm contested last year, a suit which Qualcomm won.) Qualcomm’s fastest X2E-96-100 X2 Elite Extreme runs at 4.4GHz maximum, with 18 total cores to the N1X’s 20. But Qualcomm’s chip includes all extreme cores, while the RTX Spark platform includes a mixture of beefy extreme cores and the somewhat less powerful performance cores.
Mark Hachman / Foundry
The key, for me, is single-core performance. The Snapdragon X2 Elite demonstrated single-core Geekbench 6 scores of about 4,000 in Qualcomm’s own lab, and slightly less in the field. However, Nvidia’s Grace Blackwell superchip with the Arm CPU — believed to be the foundation of the N1X CPU — scored just 3,080 on the Geekbench 6 benchmark. But the “N1X” is not expressly named, so this is just speculation.
On paper, that would give Qualcomm’s chip an advantage in code compilation, OS and browser responsiveness, and productivity tasks (such as apps like Microsoft Excel) which generally depend on the output from a single CPU core. Remember, Apple’s Mac mini is now the darling of the local-AI crowd; it’s powered by Apple’s M4 Pro at the high end. Eventually, Apple’s mini PC will likely move to the M5 Pro, too.
(Because you’ll ask: Overall, it appears that the Qualcomm Snapdragon X2 Elite Extreme scores about 3,950 on Geekbench 6’s single-core benchmark — roughly equal to the Apple M4 Pro chip within the current [2025] version of the Mac mini. That’s followed by the Nvidia Grace Blackwell chip, with a single-core score of about 3,080. Intel’s Core Ultra 300 (Panther Lake) X9 scores about 2,800, a little less than the AMD Ryzen AI Max+’s score of about 2,900 in the Framework Desktop. As for the Mac M5 Pro, its single-score comes in at about 4,350, which would place it back on top.)
To be fair, it’s unclear how much importance you should assign this. I recently tested a 7-year-old laptop versus a two 2026 models, to get a sense of how much performance has really changed over the last few years. In terms of single-core performance, the Snapdragon X2 Elite was more than twice as fast (150 vs. 62 in Cinebench 2024’s single-core test) as an old Core i7-10665G7, which only worked out to about a 2:1 advantage in an office productivity test. The two 2026 laptops differed in the time it took to calculate various Excel functions, for example, in just fractions of seconds…but that small difference would be felt if those calculations were run over and over again.
Skärmdump
The Snapdragon X2 Elite could also have an advantage in agentic AI, the new focus for everyone from Anthropic, OpenAI, and Microsoft. A paper from Intel plus public statements from AMD have characterized the CPU, not the GPU, as the controller for AI agents, potentially making it more important than the GPU. Again, the idea behind agentic AI is that those agents are constantly calculating tokens in the search for answers and performing tasks, working the CPU continually. If so, Qualcomm may have the edge.
GPU performance: Nvidia likely wins, and easily
In graphics, of course, the pendulum swings the other way. Nothing that Qualcomm has produced should be able to keep up with an RTX 5070-class GPU. If we take Nvidia’s RTX Spark core count as equivalent to an RTX 5070 Ti notebook chip, the 3Dmark Time Spy score for that GPU is about 21,000. Our Snapdragon X2 Elite test laptop scored about 4,230 using the same benchmark.
(You can see Nvidia’s Jacob Freeman playing Alan Wake 2 on a Surface Laptop Ultra via X.)
I haven’t had a chance to test the Snapdragon X2 Elite in terms of gaming, just in general CPU performance and power efficiency. But GhobsoGaming ran 20 games on top of the Snapdragon X2 Elite, and found that games were far more playable on the second-generation chip than the first — not much beyond 1080p resolution, but at about 60 frames per second on high or maximum settings.
The GPU is obviously critical for multimedia work, including photo and video editing, in addition to gaming. But the shared 128GB of embedded DRAM should also give the RTX Spark / NX1 a leg up for intensive AI-based content creation.
Remember, Nvidia has had literal decades to establish relationships with app and game developers, too. In addition to its AI-based frame generation technology for gaming — which Qualcomm lacks — you won’t find any content-creation application that hasn’t worked closely with Nvidia to optimize its code.
Windows and gaming emulation
For general productivity applications like Microsoft 365 and most of the Adobe suite, Arm compatibility has ceased being a problem — in fact, many productivity applications are coded natively for the Arm architecture. (Adobe publishes a list of exceptions; Lightroom Classic, for example, runs via Microsoft’s Prism emulation. That simply means that the application will run without issue, but that performance will be slower than if run on native code.) We’re also at a point where even more niche apps like VPNs run on Arm.
I haven’t tested everything myself, but the WindowsonArm.org database is a great place to check to see if an app runs natively or via emulation. It also maintains a list of which apps do not run on Arm, period, including AutoCAD and Acronis True Image 2025.
Gaming does invite some questions. The vast majority of games are coded for x86 processors, both the Snapdragon X2 Elite and RTX Spark will be a downgrade from the app perspective alone. Anti-cheat systems like Easy AntiCheat, Denuvo, and BattlEye have an Arm path, which has been a big stumbling block. However, older DRM systems might not always work. It’s another case of “it should work, but not always.”
In terms of AI, however, Apple has blazed the trail with its own Arm chips, the M-series silicon. CPU inferencing apps should have no issue running on either chip. Still, many inference apps have been designed to run on GPUs, specifically Nvidia’s CUDA cores. AI apps run on the Snapdragon X2 Elite will need to be specifically coded for the open ONNX format or coded and quantized for either Qualcomm’s NPU or GPU. Qualcomm’s NPU is probably the most efficient, but power users will look toward the RTX Spark, and it’s not even close. Either way, it’s a clear advantage for Nvidia.
Bottom line
As general productivity machines, then, the difference isn’t that clear-cut. For AI and gaming, however, Nvidia’s GPU offers it a distinct advantage.
It will come down to whatever applications you typically run, and whether you can live with any of the disadvantages of a particular platform. And there are the unknowns: How much power will each platform consume? And, more importantly, how much will they cost?
Asus
If you do end up wanting to buy a Snapdragon X2 Elite mini PC, the Asus Ascent QN10 is the only one I’ve seen listed. Like most PC makers, Asus is hedging certain things: We know there’s a Snapdragon X2 Elite inside, but not the specific chip, its price, or when it will ship. There will be up to 32GB of DRAM, of course, and onboard storage options between 512GB, 1TB, and 2TB (4TB total) with two PCIe M.2 slots, Wi-Fi 7, and Bluetooth 5.4 Ports abound: two USB4 and two USB-A ports on the back, plus audio; on the front, there’s another USB4 port, two USB-A ports, an HDMI 2.1 connection and a RJ45 jack. It’s all powered by a 180W charger.
Still: AMD, Intel, Nvidia, Qualcomm. That’s four distinct CPU-GPU offerings, and all four will be on the market by this fall. We’ve always cheered for more competition. By the holidays, you’ll have more to choose from in the PC market than you’ve seen in decades.
Additional reporting by Alaina Yee.
