Nvidia RTX Spark N1X in the Lenovo Yoga Pro 9i: a powerhouse with a catch
5/11/2026
Translation: machine translated
Pictures: Martin Jud
With the RTX Spark N1X, Nvidia is launching its first proprietary laptop chip. In the Lenovo Yoga Pro 9n, it becomes clear: the hardware is impressive, but the software lags behind. A review between records and reality.
Nvidia is breaking new ground. The company dominates the market for graphics processors and now wants to gain a foothold in laptop SoCs as well. With the RTX Spark N1X, it is taking on AMD, Apple, Intel, and Qualcomm. A SoC—or System-on-a-Chip—combines a CPU, GPU, and NPU on a single chip. The N1X, based on the Arm architecture, also features Unified Memory, a shared memory pool for both the CPU and GPU.
For this article, I compared the Spark in the Lenovo Yoga Pro 9n with the best laptop chips currently available from the competition. In some benchmarks, it outclasses its Windows rivals. In everyday use, however, it hits limits that have less to do with the hardware and more to do with the software.
Hardware specifications: a brief overview of the Lenovo Yoga Pro 9n and the Nvidia chip
Even though the focus of this article is on the chip, I would like to introduce the test device first.
The Yoga Pro 9n is a 15.3-inch notebook with an OLED display (2560 × 1600, 165 Hz, 1100 nits peak), a 2 TB PCIe 5.0 SSD, and 128 GB of LPDDR5X-9400 RAM. It weighs 1.75 kg and features two USB-C ports (USB4), two USB-A ports, HDMI 2.1, and an SD card reader. The Yoga feels high-quality in the hand, and the keyboard has a pleasant tactile feel. And when I open the display for the first time, I immediately notice that the OLED panel is a real feast for the eyes. The device runs Windows 11 for the Arm architecture.
The three variants of the N1X chip
Nvidia offers the RTX Spark in three configurations. The full laptop variant, which is in my test device, has 20 CPU cores. Ten of these are fast Arm Cortex-X925 cores for maximum performance, while the other ten are efficient Cortex-A725 cores for light everyday tasks. In addition, there is a Blackwell GPU with 6144 cores, an NPU, and up to 128 GB of Unified Memory. Then there is the stripped-down laptop variant, which has 18 CPU cores, 5120 GPU cores, and a maximum of 32 GB of RAM. Finally, there is the desktop variant with 20 CPU cores, 6144 GPU cores, and a 140-watt TDP instead of the 80 watts for the laptop variants. However, it does not have an NPU. Relevant for my test is the full laptop variant: it is the most powerful configuration that Nvidia currently offers for notebooks.
CPU shines in multi-core with 20 cores, but is no Apple hunter
To measure processor performance, I rely on Cinebench 2024 and Geekbench 6—the marathon and the sprint.
In the Cinebench 2024 multi-core test, the Spark achieves 1695 points with its 20 cores.
This puts Nvidia in second place—just ahead of the AMD Ryzen AI Max+ 395 and significantly ahead of Intel and Qualcomm. Only the Apple M5 Max remains out of reach.
However, when you put these numbers in relation to the number of cores, the victory against AMD is put into perspective: the Spark needs 20 cores for its 1695 points, while AMD almost reaches it with only 16 cores. And Apple outclasses the entire field with just 18 cores. The Spark can therefore keep up in multi-core tasks because it simply has the most cores in the field—not because its individual cores are particularly efficient. For a first attempt, this is still a solid result, as Nvidia clearly beats the competition from Intel and Qualcomm.
But what happens if I ignore the 20 cores and let only one work?
The single-core test shows: even without the support of the other 19 cores, the Spark is ahead of Intel and AMD. However, it falls short against the Arm leaders from Apple and Qualcomm.
Geekbench 6 paints a similar, yet slightly different picture.
In the multi-core test, the Spark achieves 19,785 points—and thus falls behind AMD. Geekbench is not a single marathon, but a series of short sprints. AMD can clock its cores more aggressively in these short bursts, while the Spark seems to be tuned for long-term stability. The Spark is currently not a sprinter, but a long-distance runner.
In the Geekbench single-core test, it also lands in third place with 3154 points, as it did in Cinebench—albeit slightly further behind Qualcomm than in the marathon.
iGPU: strong against AMD, Intel, and Qualcomm—no chance against Apple
I measure pure graphics performance with Geekbench 6 GPU—the only benchmark for which comparison values are available for all competitors. It is important for the classification that I am comparing exclusively integrated graphics solutions (iGPUs), not dedicated graphics cards. The result is a bombshell. The Spark outclasses the entire Windows competition.
This is particularly evident against AMD: the Radeon 8060S, which belongs to the same APU family as the one in the Steam Deck, was considered the non-plus-ultra of iGPUs for months. But the Spark achieves almost double the score. The Blackwell architecture with 6144 cores makes the difference here.
However, the Spark stands no chance against Apple. The M5 Max outclasses the field. Nvidia's entry is a strong statement against AMD, Intel, and Qualcomm, but it is not (yet) enough against Cupertino.
But synthetic benchmarks are one thing. Whether this performance translates into real graphics work or games is another question. The answer to that is complicated.
Gaming: between 60 FPS and a black screen
After the pure computing benchmarks, I want to know how the Spark performs in gaming. I start with 3DMark Fire Strike, which simulates a typical 1080p game scene under DirectX 11 and tests the interaction between CPU and GPU. Apple is missing here because the test is not available under macOS.
The result is sobering: the seemingly gigantic lead from the Geekbench GPU test shrinks to zero compared to AMD. Both are neck-and-neck, while Intel and Qualcomm remain far behind. The reason for Nvidia's sober result is likely that AMD's mature x86 drivers work more efficiently than the new Arm approach.
3DMark is a good indicator, but not a real game. So I start the in-game benchmark of “Cyberpunk 2077” in 1440p, with high details, without ray tracing and frame generation. And suddenly Apple is back in the game—“Cyberpunk” is one of the few AAA games available for the Mac. A small dig at Cupertino's gaming library is permitted here.
The benchmark runs at an average of 60 frames per second. This not only leaves Intel and Qualcomm far behind, but also beats the AMD competitor by 20 percent. 60 FPS in a AAA title on an Arm Windows laptop with integrated graphics—that is a statement. Nevertheless, the crown here also belongs to Apple.
Source: Screenshot / Martin Jud
Shortly thereafter, the euphoria when starting “Red Dead Redemption 2” comes to an abrupt end. The launcher opens, a loading bar races from left to right when the game starts, a first image flashes—and the game crashes. The whole thing is accompanied by an error message: “Your computer does not meet the minimum recommended requirements for the graphics card's video memory.”
Source: Screenshot / Martin Jud
I am sitting in front of a notebook with 128 GB of RAM, which also serves as video memory, and a game from 2018 tells me that is not enough. The cause is likely a bug in the Vulkan interface under Windows on Arm.
The hardware is ready for gaming, but the ecosystem is not. This is evident not only when starting games, but also during gameplay itself. In other tests with the N1X, frame drops occur. In “Age of Empires IV”, a 99th-percentile frame time of almost 78 ms was measured. In the worst moments, the image takes almost five times as long as would be necessary for smooth 60 FPS. The good average masks such stutters.
For me, this means: anyone who wants a laptop for occasional gaming is currently better off sticking with AMD, despite the impressive Cyberpunk numbers. Not because the hardware is better, but because x86 games simply run.
NPU: theoretically strong, practically (not yet) measured
A look at the official data sheet shows that Nvidia has installed an NPU in the N1X, i.e., a Neural Processing Unit for AI tasks. However, an isolated performance figure for this NPU is missing. Under the heading “AI Performance”, Nvidia only mentions a total performance of 1 petaflop, which is composed of the AI computing power of the CPU, GPU, and NPU. How strong the NPU is on its own remains in the dark.
What this NPU brings to everyday life can still be explained. It is not a replacement for the CPU or GPU, but a supplement for specialized continuous tasks such as live subtitles or background blur. It handles these tasks with significantly less energy and heat consumption. This saves battery life and keeps the fans quiet—a tangible advantage for a notebook.
How good the Spark's NPU really is cannot currently be measured. The common benchmark Geekbench AI does not yet support the NPU of the N1X.
Energy management: despite a large battery, less runtime than the competition
The Yoga Pro 9n contains a 92.5 Wh battery. This gives the notebook almost double the capacity of many competitors. You would think that would easily be enough for a long day without a power outlet. To see how much of that remains during light use, I performed the video playback test from PCMark 10. This involves playing a video at a standardized 200 nits screen brightness until the battery is empty.
The Yoga lasts 12 hours and 53 minutes. That sounds like a lot at first, but it is a bitter disappointment. For example, the Microsoft Surface Laptop with the Qualcomm Snapdragon X2 Elite X2E-78-100 and a significantly smaller 54 Wh battery reaches 19 hours and 28 minutes. Despite having almost double the battery capacity, the Nvidia chip lasts significantly shorter than the competition.
The reason for this is likely the energy management, because the N1X apparently does not clock down as low as the Snapdragon when idling and during light tasks. Software maturity therefore also lags behind the hardware here.
Conclusion: a first step, not yet a game changer
The RTX Spark N1X impressively shows what Nvidia's hardware is capable of. In terms of the CPU, it proves to be a solid long-distance runner; in terms of the iGPU, it outclasses the entire Windows competition; and in “Cyberpunk 2077”, it proves that 60 FPS in a AAA title on an Arm laptop with Windows is possible.
But the software keeps pulling the emergency brake. It starts with the games. “Red Dead Redemption 2” crashes on startup, and in other games, frame times get out of hand despite a good average. It continues with power consumption, where the huge battery still delivers less runtime than the competition. And it culminates in the realization that the x86 competition from AMD simply works in practice, while Nvidia still has to fight against the pitfalls of Windows on Arm and emulation.
For users who also want to game occasionally, the Spark is not yet a recommendation. For content creators, things look different. Anyone who benefits from the 128 GB of Unified Memory and the powerful iGPU gets an exciting glimpse into the potential of the chip here.
The Spark is therefore not yet a game changer, but it is the first serious proof that Nvidia's data center DNA can work on a laptop. If Microsoft, Nvidia, and game developers get a handle on the ecosystem, this first step could soon become a major attack. Anyone willing to put up with the teething troubles gets a piece of the future today.
Martin Jud
Senior Editor
martin.jud@digitecgalaxus.chI find my muse in everything. When I don’t, I draw inspiration from daydreaming. After all, if you dream, you don’t sleep through life.
Background information
Interesting facts about products, behind-the-scenes looks at manufacturers and deep-dives on interesting people.
Show allThese articles might also interest you
Show all
These articles might also interest you
25 comments
later