Prototype Surface Laptop Reveals Nvidia RTX Spark Limits
A TechPowerUp forum user published benchmarks of an alleged Microsoft Surface Laptop Ultra engineering prototype using Nvidia's RTX Spark N1X chip ahead of the platform's reported launch. The 24GB unified-memory unit produced respectable CPU benchmark scores but showed unstable software behavior, failed CUDA tests, high CPU temperatures, and higher idle power after a preview-driver update, according to TechPowerUp and VideoCardz.
A TechPowerUp forum user published a month of testing from an alleged Microsoft Surface Laptop Ultra engineering prototype powered by Nvidia's unreleased RTX Spark N1X system-on-chip. The user, Fouquin, reported finding the device near Microsoft's Redmond headquarters, but the unit's provenance has not been independently verified. Multiple outlets characterize it as an EV1.5 pre-production system, so its results should not be treated as retail performance.
According to TechPowerUp, the prototype initially ran an older graphics driver before Fouquin updated it to Nvidia's 616.00 developer-preview driver, which added CUDA support for the integrated Blackwell GPU. The report describes a device with 24GB of unified memory and a 512GB SSD, rather than the maximum configurations previously reported for the platform.
Hardware and benchmark results
VideoCardz and HWBusters report that the tested N1X configuration contains a 20-core Arm CPU and a Blackwell GPU with 6,144 CUDA cores. The reported CPU benchmark results were:
- •Cinebench 2024: 1,386 multi-core and 123 single-core points.
- •Cinebench 2026: 5,771 multi-core and 540 single-core points.
- •Power limits: testing used an 80W PL1 and 95W PL2 configuration, according to VideoCardz.
HWBusters reported that, in Balanced mode, the CPU settled around 35W to 37W and roughly 2.6GHz after reaching 98C to 100C. In Best Performance mode, it reportedly used 44W to 50W at about 2.8GHz while still reaching 100C. The GPU reportedly ran between 80C and 88C, with clock speeds around 2.0GHz to 2.2GHz in some tests; HWBusters described a nominal 50W GPU power limit under Nvidia Max-Q behavior.
Driver and AI workload issues
The software stack was the more consequential limitation in the published testing. TechPowerUp reported that CUDA workloads in the Phoronix Test Suite did not complete, leaving Vulkan Compute and CPU BLAS results available instead. VideoCardz likewise reported that the AI suite failed to complete CUDA tests and that shifting from Balanced to Best Performance did not produce measurable AI-test gains.
According to TechPowerUp, the 616.00 preview driver raised idle draw from roughly 7W to 8W in the earlier Balanced configuration to 12W in Balanced mode and 23W in Best Performance mode. The report also described a repeatable lock-up after about an hour of use. VideoCardz reported game-clock variation between 1.5GHz and 2.3GHz and periodic stutter.
TechSpot characterized the benchmark picture as behind Apple's M5 Pro in its cited tests, while noting that the system reportedly delivered similar performance on battery and external power. Because the results come from one unverified prototype, pre-release firmware, and developer drivers, cross-platform comparisons remain provisional.
For ML developers, the failed CUDA runs are more important than the headline compute specifications. A laptop with unified memory, an Arm CPU, and a Blackwell-class GPU could be relevant for local inference and CUDA development, but comparable early Windows-on-Arm platforms commonly depend on driver maturity, framework support, and reliable power-management behavior before benchmark potential translates into reproducible workflows.
Key Points
- 1An alleged Surface Laptop Ultra prototype exposes early RTX Spark N1X benchmark data, but pre-production hardware limits comparison with retail systems.
- 2CUDA tests in Phoronix did not complete, making software compatibility and driver stability the immediate concern for local AI workloads.
- 3Reported unified-memory hardware is promising for portable inference, while similar platforms typically require mature drivers before dependable developer deployment.
Scoring Rationale
The report provides unusually detailed pre-release data on Nvidia's prospective Arm-based laptop platform and its CUDA software stack. Its practical relevance is constrained by the unverified prototype, non-final drivers, and incomplete AI benchmark runs, but it is notable for developers tracking local AI hardware on Windows on Arm.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
