ASUS ProArt P16 H7607BA Review: RTX Spark N1X, 128GB Unified Memory and Windows on Arm Gaming

Independent reviews from GeekaWhat and NAG give us a clear, tested look at how NVIDIA’s consumer Windows-on-Arm debut actually behaves in the wild. Whether you are a developer eyeing local large language models (LLMs), a video editor married to an OLED display, or a PC gamer wondering if Microsoft’s Prism emulation layer can really handle modern titles without face-planting, the ProArt P16 H7607BA is a fascinating—and expensive—test case for the future of Windows 11 hardware.

Inside the Silicon: What Exactly Is NVIDIA RTX Spark N1X?​

If the name "Spark" sounds familiar to enterprise and workstation watchers, it should. As GeekaWhat notes, the platform shares clear DNA with NVIDIA’s compact Linux-based DGX Spark desktop box (built around the GB10 Grace Blackwell chip), now scaled down into a 45W to 80W TDP laptop envelope running Windows 11.

Rather than pairing an Intel or AMD processor with a separate PCIe graphics card—as previous ProArt P16 generations did—the RTX Spark N1X integrates the CPU, GPU, and memory controller into one unified package linked via NVIDIA's NVLink-C2C interconnect. According to NVIDIA’s official specifications and developer documentation, the N1X comes in two mobile configurations (confusingly sharing the exact same "N1X" model name, a branding quirk GeekaWhat rightly criticizes):

• Flagship 20-Core / 6,144-CUDA-Core Tier: Features a 20-core Arm v9.2 Grace CPU arranged in two 10-core clusters (each pairing five Cortex-X925 performance cores with five Cortex-A725 efficiency cores), an integrated 6,144-core Blackwell RTX GPU, 1 Petaflop of AI performance, and support for up to 128GB of unified LPDDR5X memory.
• 18-Core / 5,120-CUDA-Core Tier: Features an 18-core Grace CPU, a 5,120-core Blackwell RTX GPU, 827 Teraflops of AI performance, and support for up to 32GB of unified LPDDR5X memory.

Both variants include fourth-generation Ray Tracing cores, fifth-generation Tensor cores, an NVIDIA NPU, native CUDA 12.1 support, and compatibility with DLSS 5, Frame Generation, and both Game Ready and NVIDIA Studio drivers.

Section Summary: The RTX Spark N1X replaces the traditional CPU-plus-dGPU laptop layout with a single Arm v9.2 and Blackwell SoC. While the flagship chip matches a desktop RTX 5070 in raw CUDA core count (6,144 cores), mobile thermal and power constraints (45–80W TDP) put its real-world target squarely in the laptop RTX 5070 class.

The Unified Memory Advantage—and Its Windows 11 Caveats​

Why is unified memory such a big deal on a Windows PC? On a traditional x86 laptop with a discrete GeForce RTX 5070 Laptop GPU, your graphics card is fenced into its own dedicated GDDR memory pool. Once a 3D scene or local AI model exceeds that dedicated VRAM, performance falls off a cliff—regardless of whether you have 64GB of system RAM sitting in SO-DIMM slots.

On the ProArt P16 H7607BA, the CPU and Blackwell iGPU share a single coherent pool of onboard LPDDR5X memory running across a 256-bit interface at 300 GB/s of bandwidth. ASUS offers the H7607BA with 24GB, 32GB, 48GB, 64GB, or 128GB of soldered memory. According to ASUS, the 128GB configuration allows developers to run massive LLMs—up to 120 billion parameters with a one-million-token context window—entirely locally without touching the cloud.

Did that local AI promise hold up under testing? In NAG’s review of the 128GB UHD+ configuration, the publication tested the 27-billion-parameter Qwen3.8 model locally using Ollama. Across three short-prompt runs, the ProArt P16 achieved a median generation speed of 25.41 tokens per second (roughly 19 English words per second), alongside Geekbench 7 scores of 2,690 single-core, 22,915 multi-core, and 184,857 in CUDA compute. Because CUDA runs natively on the RTX Spark platform, AI developers do not have to wrestle with translation layers for their core compute stacks.

However, veteran Windows administrators and developers should keep two architectural realities in mind based on NVIDIA’s Windows on Arm Porting Guide:

1. Shared does not mean 100% available to the GPU: Windows 11 divides unified memory into logical regions, reserving specific budgets for the GPU and managing shared versus CPU-only memory. NVIDIA explicitly warns developers not to allocate the entire GPU memory budget, as starving the Grace CPU cores of memory can leave the operating system unresponsive.
2. Zero upgradeability: Unlike conventional gaming laptops such as the Gigabyte Gaming A16—which GeekaWhat noted still provides user-upgradeable SO-DIMM slots—the ProArt P16’s LPDDR5X is soldered directly to the motherboard. Furthermore, ASUS’s spec sheet warns that due to surface-mounted components near the motherboard's M.2 slots, only single-sided M.2 2280 SSDs are supported, and the drivers strictly require Windows 11 version 24H2 or later.

Section Summary: Up to 128GB of coherent unified memory plus native CUDA makes the ProArt P16 a uniquely capable mobile workstation for local LLMs, proved out by NAG’s 25.41 tokens/sec test in Ollama. However, buyers must choose their RAM tier carefully at checkout, as the memory is permanently soldered.

Gaming on Windows on Arm: RTX Spark vs. Laptop RTX 5070​

For years, "gaming on Windows on Arm" was a punchline involving stuttery frame rates, broken shaders, and aggressive anti-cheat blocks. With NVIDIA stepping in alongside Microsoft’s Prism x86-to-Arm64 translation layer, how does a 64GB ProArt P16 (with the 3K OLED panel) stack up against a conventional x86 gaming machine like the Gigabyte Gaming A16 (equipped with an AMD Ryzen 7 260 and a discrete GeForce RTX 5070 Laptop GPU)?

GeekaWhat put both laptops through a head-to-head gaming gauntlet, and the results show a remarkably tight race at native resolutions, with the discrete RTX 5070 laptop tending to pull ahead when leaning on DLSS and Frame Generation:

• Fortnite (Native Arm64): At 1440p High, the ProArt P16 beat the RTX 5070 laptop with an average of 70.3 FPS vs. 62.9 FPS. However, dropping to 1080p competitive settings swung the lead decisively to the x86 RTX 5070 laptop, as lower resolutions shift the bottleneck onto the CPU within the shared SoC thermal envelope.
• Pragmata: At 1440p High without upscaling, the P16 edged ahead at 67 FPS (roughly 1.3 FPS faster than the 5070 laptop). Turning on DLSS Quality put the RTX 5070 laptop about 7 FPS ahead, though adding ray tracing narrowed the 5070's lead to around 4 FPS.
• Doom: The Dark Ages: Practically a dead heat at 1440p Medium, with the P16 hitting 40 FPS (0.7 FPS ahead of the 5070 laptop). Enabling DLSS Balanced gave the RTX 5070 laptop a ~7 FPS lead, which widened to ~14 FPS with 3x Frame Generation.
• Gears of War: E-Day: At Medium settings, the P16 trailed by just over two frames (37.9 FPS vs. 40.1 FPS), maintaining a nearly identical gap with DLSS enabled (46.7 FPS vs. 48.4 FPS).
• 007 First Light: The P16 led at 1440p High without upscaling, dipped slightly behind with DLSS Balanced, and then surged to a ~17 FPS lead over the RTX 5070 laptop when using 3x Frame Generation.
• Cyberpunk 2077: GeekaWhat found this to be the discrete RTX 5070 laptop's strongest showing at High settings (56 FPS vs. 46 FPS on the P16). When switched to Ray Tracing Low with DLSS Balanced, the P16 briefly nudged ahead, before Frame Generation put the 5070 laptop back in front by about 5 FPS. Separately, NAG tested Cyberpunk 2077 on its 128GB P16 unit using ray tracing, DLSS Ultra Performance, and 4x Frame Generation, recording 102.32 FPS (alongside a 142 FPS average in a 30-minute 1080p FurMark stress test where the GPU stabilized at 85°C).

What Is the Real Cost of Prism Emulation?​

Because most PC games are still compiled for x86_64, GeekaWhat isolated the translation penalty by testing Marvel Rivals, which offers both an x86 build (run via Microsoft Prism) and a native Arm64 build:

Crucially, GeekaWhat reported zero crashes or bizarre emulation glitches across its entire test suite. Furthermore, to address Windows on Arm’s historical Achilles' heel—kernel-level anti-cheat—NVIDIA states that support for Epic’s Easy Anti-Cheat and EA’s Javelin is on the way, alongside native Arm builds or emulation optimizations for titles including Fortnite, Alan Wake 2, Cyberpunk 2077, PUBG, and upcoming Call of Duty releases.

Section Summary: In raw rasterization at 1440p, the ProArt P16 goes toe-to-toe with a conventional RTX 5070 gaming laptop while running cooler and quieter. While x86 emulation via Prism incurs a measurable FPS tax in synthetic benchmarks and high-framerate Frame Generation scenarios, live gameplay in Marvel Rivals without upscaling was virtually indistinguishable (51 FPS vs. 50.5 FPS).

Chassis, Display, and Day-to-Day Creator Trade-Offs​

Numbers on a chart only tell half the story. The ProArt P16 H7607BA weighs just 1.77 kg (3.90 lbs) and measures 1.29 to 1.59 cm thick—making it dramatically slimmer, lighter, and quieter than a plastic-bodied gaming notebook like the Gigabyte A16.

ASUS equips the laptop with a 16:10 ASUS Lumina OLED touchscreen (available in a 2880×1800 3K panel rated up to 1,100 nits HDR peak or a 3840×2400 UHD+ panel rated up to 1,600 nits HDR peak, both running at 120Hz with 100% DCI-P3 coverage and stylus support). Connectivity is equally generous for creators on the move: three USB4 Type-C ports (40Gbps with DisplayPort and Power Delivery), one USB 3.2 Gen 2 Type-A port, a full-size HDMI 2.1 FRL port, an SD Express 7.0 card reader, and Wi-Fi 7. Creative apps from Adobe (Photoshop, Premiere), Blackmagic Design, Blender, CapCut, ComfyUI, and OTOY are supported from day one, along with ASUS’s bundled MuseTree and StoryCube utilities.

Still, neither review treats the ProArt P16 as flawless:

1. The Price Tag: NAG listed its 128GB review configuration at roughly R116,000, and GeekaWhat emphasized that RTX Spark carries an early-adopter premium.
2. Physical Quirks: While GeekaWhat praised how close the chassis gets to Apple’s M5 Pro MacBook Pro 16-inch, it noted that Apple’s aluminum unibody still holds a slight edge in overall fit and finish, that the P16’s screen hinge felt slightly less convincing when poking at the touchscreen, and that MacBook converts may need time to adjust to the keyboard layout.
3. Storage Footprint: As NAG pointed out, if your configuration ships with a 1TB SSD, simultaneously storing raw 4K footage, modern 100GB+ PC games, and multi-gigabyte local LLM weights will fill the drive in short order.

The Verdict: Who Should Actually Buy the RTX Spark ProArt P16?​

If your sole mission in life is squeezing out the highest possible frame rates per dollar in Cyberpunk 2077 or competitive 1080p shooters, stop right here: a conventional x86 gaming laptop with a discrete RTX 5070 costs less, handles high-refresh 1080p CPU loads better, and lets you upgrade your RAM down the road.

Where the ASUS ProArt P16 (H7607BA) carves out its own lane is workload convergence. Before RTX Spark, no Windows-on-Arm laptop could credibly run native CUDA workflows, load colossal local AI models into a 64GB or 128GB unified memory pool, grade HDR video on a 120Hz Pantone-validated OLED display, and then casually knock out 60+ FPS in modern 1440p games inside a sub-4-pound chassis. It does not unilaterally dethrone the MacBook Pro on build polish, nor does it make x86 gaming rigs obsolete—but it proves that Windows on Arm finally has heavy-hitting silicon worthy of serious power users.