Overview of the RTX Spark Platform
Nvidia has introduced its new RTX Spark chips, a platform that was recently tested and compared against major competitors. The analysis measured performance across several metrics, including CPU capacity, GPU capacity, gaming performance, artificial intelligence (AI) performance, and overall efficiency. The comparison included chips from AMD (specifically Strix Halo), Intel, Qualcomm, and Apple‘s M5-series processors.
RTX Spark N1X Technical Details
The RTX Spark N1X is Nvidia’s initial System-on-Chip (SoC) design. It integrates a custom-built ARM CPU with Nvidia’s Blackwell GPU, supporting up to 128 GB of memory. This architectural approach is similar to those used by Apple’s M-series chips and AMD’s Strix Halo SoCs. Because the N1X utilizes an ARM CPU, compatibility requires the ARM version of Windows.
The chip has a maximum Thermal Design Power (TDP) of 80 Watts for mobile applications, while desktop systems can draw up to 120 Watts. The CPU component is a custom MediaTek CPU featuring up to 20 cores organized into four clusters. It includes ten performance cores (ARM Cortex X925, running up to 4.1 GHz) and ten efficiency cores (ARM Cortex A725, running up to 2.8 GHz). This component is manufactured by TSMC using a 3 nm process. The CPU tile is equipped with a memory interface for up to 128 GB LPDDR5x-9400 RAM, display connectors (HDMI 2.1a & DisplayPort 2.0), and high-speed I/O connectivity, such as 4x USB 4.0, PCIe 5.0, and Wi-Fi 7.
The GPU component is a Blackwell GPU (GB20B), featuring either 5120 or 6144 CUDA cores, which can operate up to 2535 MHz. Since the chip generation mirrors current consumer GPUs like the GeForce RTX 5070, the N1X GPU can benefit from existing software optimizations such as DLSS5 or Frame Generation.
The maximum memory bandwidth reaches 300 GB/s, a figure comparable to the Apple M5 Pro SoC. Both the CPU and GPU components are connected via NVLink, providing a bandwidth of 600 GB/s. In addition to the full-featured N1X (20 CPU cores and 6144 GPU cores), a smaller N1X configuration is available, featuring 18 CPU cores (9x performance + 9x efficiency) and 5120 GPU cores. Due to the identical naming convention, purchasers must verify the precise specifications when acquiring a laptop with RTX Spark.
Devices Under Test
The testing was performed on the updated 2026 Asus ProArt P16, which was specifically redesigned for the RTX Spark chips. This review unit utilized the high-end RTX Spark N1X chip, featuring 20 CPU cores, 6144 GPU cores, and 64 GB of RAM (LPDDR5x-9400), paired with a 3K OLED display. The chip consumes up to 85/70 Watts in Performance mode. Monitoring of CPU and GPU consumption data is currently limited to Nvidia’s performance overlay (Alt+R).
Furthermore, the Microsoft Surface Laptop Ultra was also reviewed, equipped with the faster RTX Spark N1X chip (20 CPU cores, 6144 CUDA cores) and 64 GB of RAM. The Surface Laptop Ultra also runs the N1X at the full 80 Watts, resulting in performance very similar to the ProArt P16. The full reviews for both machines are scheduled for release soon.
Performance Analysis and Benchmarks
Single-Core Performance and Efficiency
In single-core workloads, the N1X performed strongly, surpassing AMD’s Zen 5 chips and matching Intel’s latest Panther Lake chips. Qualcomm’s new X2 chips were noted as slightly faster, while Apple maintained a distinctly superior position. The efficiency ratings confirmed this trend:
| Chip | Performance Rating: Cinebench 2024 Single Core | Geekbench 6.7: Single-Core Performance rating |
|---|---|---|
| Apple M5 Max (MacBook Pro 16 2026) | 99.5 pt | 4336 Points |
| Apple M5 Pro (MacBook Pro 16 2026) | 99.5 pt | 4295 Points |
| Qualcomm Snapdragon X2 Elite Extreme (Zenbook A16 UX3607) | 151 Points | 3826 Points |
| Nvidia RTX Spark N1x (Surface Laptop Ultra 2026) | 131 Points | 3096 Points |
| Nvidia RTX Spark N1x (ProArt P16 H7607) | 130 Points | 3092 Points |
Regarding single-core power efficiency (using Cinebench 2024): the N1X achieved 4.26 Points per Watt (ProArt P16 H7607), which was superior to all AMD competitors. However, the Snapdragon X2 chips and Apple M5 chips demonstrated higher efficiency.
Multi-Core Performance and Efficiency
The N1X performed well in multi-core tasks, keeping pace with the fastest AMD Strix Halo systems. While Intel’s Panther Lake systems were easily outperformed, and most Snapdragon X2 Elite chips were also surpassed, the N1X could only slightly edge out the latest M6 chip, falling behind the M4/M5 Pro & Max models. Multi-core power efficiency was also measured, with the N1X achieving 13.6 Points per Watt (ProArt P16 H7607).
Graphics Processing Unit (GPU) Performance
The Blackwell GPU is comparable to high-end consumer GeForce cards. For benchmarks like 3DMark Steel Nomad, the N1X’s performance was comparable to powerful versions of the Nvidia GeForce RTX 5070 and approximately 50% faster than the Radeon 8060S. In this area, Intel and Qualcomm processors could not compete. Apple’s M5 Pro GPU was slower (-28%), while the M5 Max GPU was faster (+27%).
Gaming Capabilities and Efficiency
Gaming is a key aspect, and the Blackwell GPU adds interest, especially considering the ARM version of Windows. While testing revealed that some titles (such as F1 25 or Battlefield 6) may not be compatible with Snapdragon laptops, the performance was tested using Cyberpunk 2077 and Baldur’s Gate 3. In a test of Marvel Rivals at 1080p, the N1X showed an approximate 20% performance advantage, along with significantly better 1% low frame rates (9 vs. 29 fps). Although the N1X showed a clear advantage in large memory scenarios, dedicated gaming laptops with GeForce RTX GPUs were still superior overall. In the Cyberpunk 2077 Ultra Preset test, the N1X achieved 76.6 fps, placing it above the AMD Radeon 8060S (49.4 fps) and Apple M5 Pro (33.4 fps).
Artificial Intelligence (AI) Performance
Local AI is a significant focus for chip manufacturers. The capability to run models locally using ample memory eliminates the need for cloud services. For benchmarking, the analysis relied on MLPerf 2.0, as Geekbench AI was limited to CPU testing for the N1X. The MLPerf results showed that the N1X’s performance relative to the M5 Pro varied depending on the specific test.
Overall Assessment
The Nvidia RTX Spark N1X represents a solid initial effort utilizing high-end hardware. It presents a highly compelling package for users engaged in content creation and AI tasks, while also accommodating gaming. Although the CPU performance is robust for a first-generation product, efficiency still offers room for improvement. The GPU performance is respectable, matching the raw power of the GeForce RTX 5070, and benefits from the advantage of greater RAM for demanding tasks like modeling, video, and photo editing.
A primary limitation remains the need for the ARM version of Windows, which still presents compatibility issues, particularly with games. Nevertheless, the overall situation represents a marked improvement compared to first-generation Snapdragon Windows laptops. While dedicated gaming laptops with GeForce RTX GPUs remain superior for pure gaming, the RTX Spark architecture may represent the next evolution for gaming machines. The N1X is currently regarded as the best way to acquire premium Windows hardware.
| Feature | Performance vs. Competitors | Key Takeaway |
|---|---|---|
| CPU Performance | Solid, competitive with AMD Strix Halo. | Good for a first-generation product. |
| GPU Performance | On par with GeForce RTX 5070. | Strong raw power, benefits from large memory capacity. |
| Efficiency | Needs improvement, but competitive. | Apple and Snapdragon X2 chips remain the efficiency leaders. |
