Performance
RTX Spark Benchmarks
Compiling all known performance data, specifications, and hands-on impressions of NVIDIA's RTX Spark platform as reported from Computex 2026. Independent third-party benchmarks will be added when review units ship.
Geekbench 6 · CPU
Geekbench 6 Scores
| Processor | Single-Core | Multi-Core |
|---|---|---|
| NVIDIA N1x (RTX Spark ES) | 3,096 | 18,837 |
| Apple M3 Max (16" MBP, Nov 2023) | 3,124 | 18,920 |
| Apple M5 (14" MBP, 2025) | 4,224 +36% | 17,465 −7% |
| Apple M5 Pro (14"/16" MBP, 2026) | 4,242 +37% | 25,800 +37% |
⚠ Engineering Sample Caveats
Pre-release silicon · lower clock speeds than final production · Linux environment, not Windows · firmware/drivers unoptimized · final RTX Spark expected to score higher. Compared against Apple's 2023 M3 Max — Apple has since shipped M4 (2024) and M5 (2025).
Sources: Geekbench Browser (N1x result) · Wccftech · Geekbench Browser (M3 Max)
RTX Spark Specifications
| Component | RTX Spark N1 | RTX Spark N1X | Notes |
|---|---|---|---|
| CPU | 20-core Arm (Grace) | 20-core Arm (Grace) | Custom NVIDIA-designed Arm cores; perf + efficiency mix |
| GPU | Blackwell RTX | Blackwell RTX | RTX 5070-class; up to 6,144 CUDA cores |
| AI Engine | Integrated NPU | Integrated NPU | Dedicated neural processing unit on-die |
| Memory | Up to 64 GB | Up to 128 GB | LPDDR5X unified memory |
| Interconnect | NVLink-C2C (600 GB/s) | High-speed CPU↔GPU fabric | |
| AI Performance | Up to 1,000 TFLOPS (FP4) | 1 petaflop-class AI throughput | |
| Local LLM | Up to 120B params | With 1M token context window | |
| Gaming API | DirectX 12 Ultimate, Vulkan | Full ray tracing, DLSS 4.5, Multi Frame Gen | |
| Process Node | TSMC (advanced) | Specific node undisclosed | |
Source: ThePCEnthusiast, Tom's Guide, NVIDIA Computex 2026 keynote. Specifications are preliminary — final may vary by OEM implementation.
Real-World Benchmarks (Preliminary)
Early benchmark data is emerging from Computex 2026 hands-on sessions and pre-release testing. These are preliminary numbers — final review units will provide more comprehensive data. All scores should be treated as directional until independent labs publish verified results.
Developer Workload & CPU Benchmarks
Wccftech — June 2, 2026
Wccftech — June 2, 2026
Mezha — June 2, 2026
Digital Foundry — Computex 2026
Gaming Benchmarks (On-Stage Demos)
NVIDIA demonstrated RTX Spark laptops running AAA games on battery power at Computex 2026. These are controlled demos, not independent testing — but they establish a performance floor.
| Game | Settings | Observed Performance | Source |
|---|---|---|---|
| Forza Horizon 6 | High settings, 1440p | Smooth 60+ fps on battery | IGN, Wccftech |
| 007 First Light | High settings, ray tracing on | Playable framerate on battery | IGN, Wccftech |
| Alan Wake 2 | Ray Reconstruction enabled | Playable with full RT features | Wccftech |
On-stage demos from Computex 2026. These are NVIDIA-controlled demonstrations shown to press — not independent benchmarks. All games ran on battery power, demonstrating the platform's efficiency. Independent fps testing will be conducted when review units ship in H2 2026.
AI & NPU Performance
RTX Spark's AI capability is its defining differentiator. While competitors ship NPUs in the 10–50 TOPS range, RTX Spark delivers over 1,000 TOPS (FP4) by leveraging its integrated Blackwell GPU for AI inference — two orders of magnitude beyond current-generation AI PC chips.
| Chip | AI TOPS (NPU only) | AI TOPS (GPU+NPU) | Max Local LLM |
|---|---|---|---|
| NVIDIA RTX Spark | Undisclosed | ~1,000 (FP4) | 120B params |
| Intel Core Ultra (Lunar Lake) | 48 TOPS | ~100 TOPS | ~10B params |
| AMD Ryzen AI (Strix Halo) | 50 TOPS | ~120 TOPS | ~14B params |
| Qualcomm Snapdragon X Elite | 45 TOPS | ~45 TOPS | ~7B params |
| Apple M6 | 38 TOPS (ANE) | ~200 TOPS | ~32B params |
Sources: ThePCEnthusiast, Technology Org. Competitor NPU TOPS from manufacturer spec sheets. GPU+NPU figures for RTX Spark are FP4; competitors are FP16/INT8 — not directly comparable but order-of-magnitude illustrative.
Creative App Benchmarks (Preliminary)
NVIDIA has partnered with Adobe to optimize key creative applications for RTX Spark. While independent benchmarks aren't available yet, early claims and hands-on impressions provide directional performance indicators.
Hands-On Impression — Tom's Guide
"My daily driver is the M5 MacBook Pro for its lightning fast performance... but I have to carry a Steam Deck for the gaming side. Enter RTX Spark, and that compromise is gone."
Tom's Guide — Jason England, June 2, 2026. Hands-on testing of RTX Spark laptops for video editing, gaming, and AI workloads at Computex 2026. Conclusion: MacBook Pro dominance is "in trouble."
Gaming Performance
RTX Spark is the first Arm-based Windows chip with a genuinely capable integrated GPU. The 6,144 CUDA cores match the desktop RTX 5070, and DLSS 4.5 with Multi Frame Generation is exclusive to NVIDIA hardware — no competing Arm SoC offers comparable gaming features.
Gaming Features Unique to RTX Spark
What We Know — Gaming
- ⚠ No independent fps benchmarks yet. Review units expected closer to H2 2026 launch. All current data is from NVIDIA's controlled demos and Computex hands-on sessions.
- + AAA game studios are building native Arm builds. Per Upcomer, several studios confirmed working on Arm-native game ports for RTX Spark launch.
- + x86 emulation for legacy games. Windows on Arm emulation layer handles existing x86 games, though performance varies. Anti-cheat compatibility remains an open question per XDA Developers.
- + Handheld potential. PCWorld's Michael Crider argues RTX Spark could be "game-changing" for PC gaming handhelds — matching desktop RTX 5070 in a mobile SoC form factor with DLSS 4.5.
Sources: PCWorld, XDA Developers, Phoronix, Upcomer
Where RTX Spark Stands
RTX Spark enters a market with established competitors across x86 and Arm architectures. Here's how the performance picture shapes up based on announced specifications.
| Capability | RTX Spark | Apple M6 | Intel Core Ultra | AMD Ryzen AI | Snapdragon X |
|---|---|---|---|---|---|
| CPU Architecture | Arm (Grace) | Arm (Apple) | x86 | x86 | Arm (Oryon) |
| GPU Class | RTX 5070-class | ~20-core Apple GPU | Arc Xe2 | RDNA 3.5 | Adreno X1 |
| AI TOPS (GPU+NPU) | ~1,000 | ~200 | ~100 | ~120 | ~45 |
| Max Memory | 128 GB (unified) | 48 GB (unified) | 64 GB (DDR5) | 64 GB (DDR5) | 32 GB (LPDDR5X) |
| Gaming Upscaling | DLSS 4.5 + MFG | MetalFX | XeSS | FSR 4 | FSR (via Adreno) |
| Ray Tracing | Hardware RT | Hardware RT | Limited RT | Limited RT | No hardware RT |
| Local LLM (max) | 120B params | ~32B params | ~10B params | ~14B params | ~7B params |
Analysis: The Three Key Advantages
- Integrated RTX 5070-class GPU. No other SoC comes close to RTX Spark's graphics capability. Qualcomm's Adreno and Apple's GPU are 2–4× smaller in core count. Intel and AMD require discrete GPUs for this level of gaming performance — adding cost, power, and thermal complexity.
- NVIDIA's software ecosystem. CUDA, TensorRT, DLSS, and Broadcom driver maturity give RTX Spark a software moat that no Arm competitor can replicate in the near term. Adobe, Autodesk, and major ISVs are already optimizing for the platform.
- Unified memory architecture. 128 GB of unified LPDDR5X with 600 GB/s bandwidth eliminates the CPU↔GPU memory copy bottleneck. This enables local AI model sizes (120B params) that no other laptop chip can approach, and benefits creative workloads that juggle large assets between CPU and GPU memory.
Analysis synthesized from: Tom's Guide, Tom's Hardware, PCWorld, ThePCEnthusiast, Phoronix.
Open Questions
Several key performance questions won't be answered until review units ship in H2 2026. These will determine whether RTX Spark meets or exceeds the expectations set at Computex.
- ? Sustained vs. peak GPU performance. How does the RTX 5070-class GPU perform under sustained gaming loads in a thin laptop chassis? Thermal throttling could significantly reduce real-world fps from peak specs.
- ? x86 emulation overhead. Until native Arm games ship, most titles will run through Windows on Arm's x86 emulation layer. Performance penalty for emulated games could be 20–40% vs. native — negating much of the GPU advantage.
- ? Battery life under gaming load. Arm efficiency at idle is well-established, but RTX 5070-class GPU power draw under gaming could push total system power beyond what a compact laptop battery can sustain for more than 1–2 hours.
- ? Pricing tiers. NVIDIA's DGX Spark mini PC launched at $4,699. If RTX Spark laptops with 64–128 GB RAM carry premium pricing, they may compete with high-end workstations rather than mainstream laptops — limiting the addressable market.
- ? Anti-cheat compatibility. Many popular competitive games (Valorant, Fortnite, Call of Duty) use kernel-level anti-cheat that doesn't work under x86 emulation on Arm. This could lock RTX Spark out of a major gaming segment.
Last updated: June 2, 2026. This page will be updated as independent benchmarks, reviews, and performance data become available. All specifications are preliminary and sourced from Computex 2026 coverage. Check the News page for the latest coverage.