| Test | RTX Spark (ES) | Apple M4 Max (for comparison) |
|---|---|---|
| Single-Core | 540 points | ~676 points (estimated from user reports) |
| Multi-Core | 5,771 points | Significantly higher; reports indicate the M4 Max is about 35.6% faster in multi-core |
Sources report that in single-core the RTX Spark roughly matches the Apple M3 Max, while the M4 Max leads by a wide margin . In multi-core the RTX Spark also falls well behind the M4 Max . The leak also revealed configurable power limits of 80 W PL1 and 95 W PL2, closely matching Nvidia's previously disclosed RTX Spark power envelope of 45–80 W . All figures are from an engineering sample (pre-production) — final retail performance may improve with driver and firmware optimizations.
An earlier Geekbench 6 result for the N1X (previously codenamed N1x) engineering sample, posted in June 2025 and subsequently removed but archived by multiple outlets, showed :
| Test | Nvidia N1X (ES) | Apple M3 Max | Apple M5 | Apple M5 Pro |
|---|---|---|---|---|
| Single-Core | ~3,096 | ~3,124 | ~4,224 | ~4,242 |
| Multi-Core | ~18,837 | ~18,920 | ~17,465 | ~25,800 |
The N1X roughly tied the M3 Max from late 2023 but was ~36% behind Apple's M5 in single-core and well behind the M5 Pro in multi-core . The test system was an HP development board (model HP 83A3) running Ubuntu 24.04.1 .
| Component | Spec |
|---|---|
| CPU | 20-core Arm v9.2 — 10× Cortex-X925 performance cores (up to 4.1 GHz) + 10× Cortex-A725 efficiency cores |
| GPU | Blackwell 2.0 with 48 SMs = 6,144 CUDA cores + 5th-gen Tensor Cores — roughly equivalent to a desktop RTX 5070 |
| AI Performance | 1 PFLOPS (FP4 precision) using NVFP4; capable of running 120-billion-parameter LLMs locally |
| Unified Memory | Up to 128 GB LPDDR5X-9400 at ~300 GB/s bandwidth, shared CPU+GPU |
| FP32 (GPU) | ~31 TFLOPS |
| Process Node | TSMC 3nm |
| Platform | Full CUDA stack on Windows on Arm; NVLink-C2C die-to-die interconnect |
| Co-Design | Co-developed with MediaTek |
| TDP | 45–80 W (full package) |
Nvidia officially announced RTX Spark on May 31, 2026 at Computex . Shipments begin fall 2026.
| Manufacturer | Confirmed Models | Notes |
|---|---|---|
| Microsoft | Surface Laptop Ultra (15-inch, mini LED PixelSense Ultra, 2,000-nit peak brightness) | First device seen in Cinebench leak; flagship AI PC |
| HP | HP OmniBook Ultra 16 and HP OmniBook X 14 | HP claims these will be the world's thinnest RTX Spark devices |
| Dell | Dell XPS 16 Creator Edition (and others) | Confirmed multiple designs in development |
| ASUS | ASUS ProArt P16 and ASUS ProArt P14 (creator-focused, OLED displays) | Part of initial 8 confirmed laptop designs |
| Lenovo | Lenovo Yoga Pro 9n (and others) | Confirmed at Computex |
| MSI | MSI Prestige N16 Flip AI+ | MSI representatives confirmed early partnership |
Nvidia has not announced official pricing. Estimates from analysts and press :
| Tier | Estimated Price Range |
|---|---|
| N1-based devices (lower-end) | $1,799 – $2,500 |
| N1X flagship devices (20-core) | $2,500 – $2,900 |
A Morgan Stanley analyst note (based on Computex channel checks) points to N1X machines around $2,899 and N1 models near $1,799 . These are premium AI PC price points — notably higher than typical Arm laptops today and squarely in MacBook Pro territory .