RTX Spark vs. Snapdragon X2 Elite: Which chip would you like in your AI PC?

One huge factor modified at Computex this previous week: Home windows on Arm processors grew to become the subsequent huge factor for Home windows desktops, particularly for mini PCs designed for productiveness and AI purposes. So which Arm processor is the very best wager?

Till they hit our take a look at bench, we will’t say for positive. However similar to your favourite NFL analyst can predict winners and losers throughout the preseason, we will have a look at what these chips have already achieved and begin making guesses. The contenders: Qualcomm’s Snapdragon X2 Elite Excessive, which debuted in an Asus mini PC at Computex, in addition to a number of mini PCs (together with the Floor RTX Spark Dev Field) powered by the Nvidia RTX Spark, the highly effective new child on the block.

Microsoft’s formal help of OpenClaw this previous week signifies that these mini PCs will in all probability be requested to carry out agentic duties, create AI content material, and run productiveness purposes. Since we all know how properly AMD’s Ryzen and Intel’s Core processors run in mini PCs, we’re going to largely go them over in favor of the Arm choices. On paper, my guess is that the RTX Spark would be the higher machine for AI content material creation and gaming, however Qualcomm shouldn’t be counted out; the Snapdragon X2 Elite ought to have a determined benefit basically productiveness and presumably agentic AI. Right here’s why.

CPU efficiency: Qualcomm seemingly wins

We all know as a lot concerning the Nvidia N1X / RTX Spark platform as every other chip. The high-end N1X CPU (made by Mediatek in partnership with Nvidia) consists of 10 Cortex-X925 “excessive” cores, in addition to 10 Cortex-A725 “efficiency” cores, in what’s in any other case an off-the-shelf implementation of an Arm processor. Naturally, the promoting level of the N1X/RTX Spark is its highly effective built-in GPU, with 6,144 “Blackwell RTX” cores, concerning the equal of an RTX 5070 CPU.

Qualcomm, in the meantime, designed its personal customized Oryon Gen 3 CPU cores utilizing an Arm architectural license that allowed it the liberty to create its personal solely new design. (It was this license that Arm and Qualcomm contested final 12 months, a swimsuit which Qualcomm received.) Qualcomm’s quickest X2E-96-100 X2 Elite Excessive runs at 4.4GHz most, with 18 whole cores to the N1X’s 20. However Qualcomm’s chip consists of all excessive cores, whereas the RTX Spark platform features a combination of beefy excessive cores and the considerably much less highly effective efficiency cores.

Mark Hachman / Foundry

The important thing, for me, is single-core efficiency. The Snapdragon X2 Elite demonstrated single-core Geekbench 6 scores of about 4,000 in Qualcomm’s personal lab, and barely much less within the subject. Nevertheless, Nvidia’s Grace Blackwell superchip with the Arm CPU — believed to be the muse of the N1X CPU — scored simply 3,080 on the Geekbench 6 benchmark. However the “N1X” is just not expressly named, so that is simply hypothesis.

On paper, that might give Qualcomm’s chip a bonus in code compilation, OS and browser responsiveness, and productiveness duties (equivalent to apps like Microsoft Excel) which usually depend upon the output from a single CPU core. Bear in mind, Apple’s Mac mini is now the darling of the local-AI crowd; it’s powered by Apple’s M4 Professional on the excessive finish. Finally, Apple’s mini PC will seemingly transfer to the M5 Professional, too.

(Since you’ll ask: General, it seems that the Qualcomm Snapdragon X2 Elite Excessive scores about 3,950 on Geekbench 6’s single-core benchmark — roughly equal to the Apple M4 Professional chip inside the present [2025] model of the Mac mini. That’s adopted by the Nvidia Grace Blackwell chip, with a single-core rating of about 3,080. Intel’s Core Extremely 300 (Panther Lake) X9 scores about 2,800, rather less than the AMD Ryzen AI Max+’s rating of about 2,900 within the Framework Desktop. As for the Mac M5 Professional, its single-score is available in at about 4,350, which might place it again on high.)

To be truthful, it’s unclear how a lot significance it is best to assign this. I lately examined a 7-year-old laptop computer versus a two 2026 fashions, to get a way of how a lot efficiency has actually modified over the previous couple of years. By way of single-core efficiency, the Snapdragon X2 Elite was greater than twice as quick (150 vs. 62 in Cinebench 2024’s single-core take a look at) as an previous Core i7-10665G7, which solely labored out to a few 2:1 benefit in an workplace productiveness take a look at. The 2 2026 laptops differed within the time it took to calculate varied Excel features, for instance, in simply fractions of seconds…however that small distinction could be felt if these calculations had been run again and again.

Skärmdump

The Snapdragon X2 Elite may even have a bonus in agentic AI, the brand new focus for everybody from Anthropic, OpenAI, and Microsoft. A paper from Intel plus public statements from AMD have characterised the CPU, not the GPU, because the controller for AI brokers, probably making it extra vital than the GPU. Once more, the concept behind agentic AI is that these brokers are continuously calculating tokens within the seek for solutions and performing duties, working the CPU regularly. If that’s the case, Qualcomm could have the sting.

GPU efficiency: Nvidia seemingly wins, and simply

In graphics, in fact, the pendulum swings the opposite means. Nothing that Qualcomm has produced ought to be capable to sustain with an RTX 5070-class GPU. If we take Nvidia’s RTX Spark core depend as equal to an RTX 5070 Ti pocket book chip, the 3Dmark Time Spy rating for that GPU is about 21,000. Our Snapdragon X2 Elite take a look at laptop computer scored about 4,230 utilizing the identical benchmark.

(You may see Nvidia’s Jacob Freeman enjoying Alan Wake 2 on a Floor Laptop computer Extremely through X.)

I haven’t had an opportunity to check the Snapdragon X2 Elite by way of gaming, simply basically CPU efficiency and energy effectivity. However GhobsoGaming ran 20 video games on high of the Snapdragon X2 Elite, and located that video games had been much more playable on the second-generation chip than the primary — not a lot past 1080p decision, however at about 60 frames per second on excessive or most settings.

The GPU is clearly vital for multimedia work, together with photograph and video enhancing, along with gaming. However the shared 128GB of embedded DRAM also needs to give the RTX Spark / NX1 a leg up for intensive AI-based content material creation.

Bear in mind, Nvidia has had literal a long time to determine relationships with app and sport builders, too. Along with its AI-based body era know-how for gaming — which Qualcomm lacks — you received’t discover any content-creation utility that hasn’t labored intently with Nvidia to optimize its code.

Home windows and gaming emulation

For common productiveness purposes like Microsoft 365 and a lot of the Adobe suite, Arm compatibility has ceased being an issue — in actual fact, many productiveness purposes are coded natively for the Arm structure. (Adobe publishes an inventory of exceptions; Lightroom Traditional, for instance, runs through Microsoft’s Prism emulation. That merely signifies that the applying will run with out concern, however that efficiency shall be slower than if run on native code.) We’re additionally at a degree the place much more area of interest apps like VPNs run on Arm.

I haven’t examined all the things myself, however the WindowsonArm.org database is a nice place to examine to see if an app runs natively or through emulation. It additionally maintains an inventory of which apps don’t run on Arm, interval, together with AutoCAD and Acronis True Picture 2025.

Gaming does invite some questions. The overwhelming majority of video games are coded for x86 processors, each the Snapdragon X2 Elite and RTX Spark shall be a downgrade from the app perspective alone. Anti-cheat techniques like Straightforward AntiCheat, Denuvo, and BattlEye have an Arm path, which has been an enormous stumbling block. Nevertheless, older DRM techniques may not at all times work. It’s one other case of “it ought to work, however not at all times.”

By way of AI, nonetheless, Apple has blazed the path with its personal Arm chips, the M-series silicon. CPU inferencing apps shouldn’t have any concern operating on both chip. Nonetheless, many inference apps have been designed to run on GPUs, particularly Nvidia’s CUDA cores. AI apps run on the Snapdragon X2 Elite will must be particularly coded for the open ONNX format or coded and quantized for both Qualcomm’s NPU or GPU. Qualcomm’s NPU might be probably the most environment friendly, however energy customers will look towards the RTX Spark, and it’s not even shut. Both means, it’s a transparent benefit for Nvidia.

Backside line

As common productiveness machines, then, the distinction isn’t that clear-cut. For AI and gaming, nonetheless, Nvidia’s GPU gives it a definite benefit.

It is going to come right down to no matter purposes you usually run, and whether or not you possibly can dwell with any of the disadvantages of a selected platform. And there are the unknowns: How a lot energy will every platform eat? And, extra importantly, how a lot will they price?

Asus

For those who do find yourself wanting to purchase a Snapdragon X2 Elite mini PC, the Asus Ascent QN10 is the one one I’ve seen listed. Like most PC makers, Asus is hedging sure issues: We all know there’s a Snapdragon X2 Elite inside, however not the particular chip, its value, or when it is going to ship. There shall be as much as 32GB of DRAM, in fact, and onboard storage choices between 512GB, 1TB, and 2TB (4TB whole) with two PCIe M.2 slots, Wi-Fi 7, and Bluetooth 5.4 Ports abound: two USB4 and two USB-A ports on the again, plus audio; on the entrance, there’s one other USB4 port, two USB-A ports, an HDMI 2.1 connection and a RJ45 jack. It’s all powered by a 180W charger.

Nonetheless: AMD, Intel, Nvidia, Qualcomm. That’s 4 distinct CPU-GPU choices, and all 4 shall be in the marketplace by this fall. We’ve at all times cheered for extra competitors. By the vacations, you’ll have extra to select from within the PC market than you’ve seen in a long time.

Further reporting by Alaina Yee.