A leaked RTX Spark laptop shows promise—and painful growing pains
RTX Spark benchmark results from a leaked Microsoft Surface Laptop Ultra prototype show that NVIDIA’s new Arm-based N1X processor and Blackwell GPU can already match midrange desktop chips in some tests but are currently held back by immature, unreleased GPU drivers that limit gaming, AI, and even basic power behavior across the system. A forum user, Fouquin, says they “found” an engineering sample Surface Laptop Ultra with an RTX Spark N1X chip on the roadside near Redmond and then posted an informal review after several weeks of testing. That odd origin story aside, the data matters: Cinebench 2024 scores of 1,386 multicore and 123 single-core, and Cinebench 2026 scores of 5,771 multicore and 540 single-core, put Surface Laptop Ultra performance below Apple’s M5 Pro and similarly priced x86 laptops. For a platform pitched as reinventing the AI PC, that gap is hard to ignore.
Underwhelming RTX Spark benchmark numbers point to software, not silicon
On paper, the RTX Spark N1X looks anything but weak: a 20‑core Arm CPU paired with a Blackwell GPU offering 6,144 CUDA cores, 24GB of unified memory, and a 512GB SSD should not be struggling to keep up with today’s midrange chips. Yet RTX Spark benchmark results show the Surface Laptop Ultra lagging behind Apple’s M5 Pro in the Phoronix Test Suite, Blender, and Cinebench, and behind likely x86 competitors in the same price class. The more you read the numbers, the clearer it becomes that this is a driver story. Initial testing relied on late‑2025 Game Ready 591.33 drivers that could not even enable proper gaming. A preview 616.00 driver with CUDA 13.4, updated OpenGL, and Vulkan finally turned on modern features—but without delivering the expected performance lift. As one summary puts it, this engineering sample is “far from retail‑ready,” with much of its missing performance “stemming from immature firmware and drivers rather than from the NVIDIA N1X chip itself.”
Gaming on the prototype: powerful GPU, prototype‑grade experience
For gamers, the RTX Spark prototype is a warning label. Fouquin reports GPU clocks swinging wildly between 1.5GHz and 2.3GHz, which led to “massive” frame stutters every few seconds even after updating to the 616.00 preview driver. Performance barely changed whether the system ran on battery, a 64W limit, or a 105W limit, which strongly implies conservative power limits and confused boost logic inside the unreleased GPU drivers. According to the early hands‑on testing, “gaming was one of the biggest disappointments,” with games like Helldivers 2 and some GPGPU integer tests crashing outright. CUDA 13.4 support did not fix everything either: CUDA acceleration in the Phoronix AI benchmark suite frequently timed out and fell back to Vulkan compute, where results were modest. This is not the behavior of weak hardware; it is the behavior of a platform where the power scheduler, boost states, and API layers are clearly still under construction.
The 616.00 driver proves system‑level tuning is unfinished
The most telling part of this story is what happened when NVIDIA’s 616.00 preview driver landed. It finally added CUDA support for the Blackwell GPU and more detailed telemetry in NvMon, exposing boost state and PCIe link information the older driver hid. At the same time, it quietly reshaped the system’s power profiles—downwards. On the “Best Performance” profile, Fouquin saw power draw capped between 8W and 12W with no real performance improvement, even as GPU clocks continued to fluctuate and stutters remained. Idle power got worse, too, jumping from roughly 7–8W to 12–23W depending on the power mode. That is the opposite of what users expect from a next‑generation AI laptop and a clear sign that RTX Spark’s system‑level tuning—from firmware to Windows‑on‑Arm integration—is still a work in progress, not a finished product.
Why Surface Laptop Ultra performance today should not dictate your buying decision tomorrow
It is tempting to write off RTX Spark based on this one Surface Laptop Ultra prototype, but that would be a mistake. This is an EV1.5 engineering sample running unreleased GPU drivers on a non‑final panel, firmware, and power configuration, and even the early reviewer notes that an engineering sample is “unlikely to have all the performance and stability tweaks of a commercial release.” Desktops and laptops with RTX Spark are not due until fall, and NVIDIA and Microsoft “still have plenty of time to work on further system optimizations,” meaning these results should not be treated as final. Historically, driver maturity between prototype and retail launch can transform a product—especially for a new Arm platform. For now, the sensible view is this: the RTX Spark benchmark leak exposes driver optimization gaps, not a dead‑on‑arrival architecture. The real test will be retail hardware, shipping firmware, and production‑grade NVIDIA prototype drivers graduating into stable releases.
