NVIDIA’s GeForce evangelist Jacob Freeman posted a single photograph of Alan Wake 2 running on an RTX Spark-powered notebook, confirming DLSS 4.5 and Ray Reconstruction were active. The image itself is unremarkable by modern standards — a game running on a screen. But what it omits tells a far more significant story than what it shows. No resolution. No graphics settings. No frame rate counter. No benchmark results. Just a proof-of-life shot for an Arm-based SoC that represents NVIDIA’s most ambitious push into the Windows PC processor market in over a decade.
The One Photo That Speaks Volumes — and the Data Withheld
Freeman’s post on X shows the Alan Wake 2 title screen displayed on a thin-and-light notebook running the RTX Spark SoC. The accompanying text confirms DLSS 4.5 and Ray Reconstruction are enabled. That is the entirety of the performance disclosure.
What is missing is the entire performance picture: resolution, graphical preset, and achieved frame rate are all absent.
This omission is not incidental. Alan Wake 2 remains one of the most punishing titles in the current gaming landscape, demanding full ray tracing and pushing even desktop RTX 4090 cards to around 30 fps at 4K without DLSS. The RTX 5090 itself requires DLSS 4’s multi-frame generation to reach approximately 220 fps at 4K. A thin-and-light notebook SoC demonstrating that it “runs” this game is one thing; demonstrating that it runs it well is an entirely different proposition.
NVIDIA has publicly claimed that RTX Spark can deliver “over 100 fps at 1440p with ray tracing and DLSS enabled,” but has not specified which titles, which settings, or under what conditions those figures were achieved. No independent benchmarks exist. No third-party reviews have been published. With commercial availability not expected until autumn, the only data points available are the ones NVIDIA chooses to release — and this photograph is precisely that: a controlled disclosure.
What Is RTX Spark? A 14-Year Return to Windows on Arm
The RTX Spark SoC, unveiled at Computex 2026, marks NVIDIA’s first serious attempt at a Windows-on-Arm processor since the Surface RT, which shipped with the Tegra 3 SoC back in 2012. Co-developed with MediaTek and fabricated on TSMC’s 3nm process, the RTX Spark represents a fundamentally different architectural approach from the smartphone-derived chips that preceded it.
The silicon itself combines a 20-core Grace CPU with a 6,144-CUDA-core Blackwell GPU, paired with up to 128GB of LPDDR5X unified memory. The CUDA core count matches that of the desktop GeForce RTX 5070, but the thermal design is dramatically different: NVIDIA targets approximately 80W for the entire SoC, compared to the 250W-plus of a desktop RTX 5070. That thermal constraint necessarily limits sustained performance, regardless of the core count.
The memory subsystem is where the gaming implications become clearest. The RTX 5070 desktop card uses GDDR7 on a 192-bit bus, delivering 672 GB/s of memory bandwidth. The RTX Spark’s LPDDR5X unified memory, by contrast, tops out at 300 GB/s — less than half. The total 128GB pool is enormous and valuable for AI inference and creative workloads, but bandwidth, not capacity, is the primary constraint for high-frame-rate gaming. The CPU-GPU interconnect uses NVLink C2C at 600 GB/s, which helps, but the memory bottleneck remains.
Laptops featuring the RTX Spark SoC are expected from ASUS, Dell, HP, Lenovo, MSI, and Microsoft’s Surface Laptop Ultra line, all slated for an autumn launch.
DLSS 4.5 Ray Reconstruction: The Other Computex Announcement
The same Computex event where RTX Spark debuted also brought the formal announcement of DLSS 4.5 Ray Reconstruction. This update arrives after the earlier DLSS 4.5 release in January, which introduced a transformer-model-based Super Resolution and Dynamic Multi Frame Generation but left Ray Reconstruction on the previous-generation model.
The August update replaces that older model with a second-generation transformer architecture, delivering a 35 percent increase in computational capacity and 20 percent more processing parameters. Training was conducted on a significantly larger dataset, with NVIDIA claiming improvements in lighting accuracy, temporal stability, and motion clarity. Twenty-seven titles will support the new model at launch, including Alan Wake 2, Cyberpunk 2077, DOOM: The Dark Ages, and Hogwarts Legacy.
The critical detail for users is backward compatibility: DLSS 4.5 Ray Reconstruction works on every GeForce RTX GPU from the RTX 20 series onward. In an era where NVIDIA has increasingly reserved features for the RTX 50 series, this cross-generational support is noteworthy. Blender integration is also expected, with Cycles denoising support arriving in version 5.3, scheduled for autumn.
How Does RTX Spark Handle Gaming Compatibility?
One of the primary obstacles faced by previous Windows-on-Arm implementations, particularly Qualcomm’s Snapdragon X series, was gaming compatibility. The RTX Spark approach addresses this on multiple fronts.
NVIDIA and Microsoft have secured Arm-native support for Epic’s Easy Anti-Cheat and BattlEye, two of the most widely used anti-cheat systems in online gaming. Riot Games has confirmed native Arm support for League of Legends and VALORANT. KRAFTON’s PUBG is also on the compatibility list. Native Arm versions of Alan Wake 2, Naraka: Bladepoint, and War Thunder are confirmed as well.
For x86 titles that lack native Arm builds, Microsoft’s Prism emulation layer has been optimized specifically for the RTX Spark architecture. The combination of native Arm support for major online titles and improved emulation for legacy games represents a meaningful step forward for Windows-on-Arm gaming compatibility.
But compatibility is not performance. A game running through emulation or even natively on a 300 GB/s memory subsystem at 80W total package power will not match the experience of a dedicated gaming laptop with a discrete GPU. The question is not whether RTX Spark can run Alan Wake 2 — it clearly can. The question is at what resolution, at what settings, and at what frame rate.
Where RTX Spark Actually Excels
The evidence available so far suggests that gaming, while a demonstrated capability, is not the primary design target for RTX Spark. The SoC’s architecture points toward workloads that benefit from large unified memory pools and high CPU-GPU bandwidth — specifically, local AI inference and professional creative work.
The 128GB unified memory enables execution of models up to approximately 120 billion parameters entirely on-device, a capability that has no direct competition in the thin-and-light notebook segment. Adobe has announced dedicated optimizations for Photoshop and Premiere. 3D scene rendering workloads exceeding 90GB fit comfortably within the memory envelope. These are the use cases where the RTX Spark’s architectural decisions make unambiguous sense.
Gaming compatibility, in this context, functions as a proof point — a demonstration that the platform is not a single-purpose AI appliance but a general-purpose Windows PC that can also play modern titles. That is a meaningful distinction for buyers who want one device for work and play, but it is not the same as positioning RTX Spark as a gaming-first platform.
Until independent benchmarks are published at launch, the prudent stance is to treat NVIDIA’s gaming performance claims as unverified. The single photograph of Alan Wake 2 running on an RTX Spark notebook confirms one thing: the game boots and displays correctly with DLSS 4.5 and Ray Reconstruction active. That is a technical achievement for an Arm SoC at 80W. It is not, however, a performance validation.
Autumn will bring shipping hardware, independent reviews, and real-world benchmarks. Until then, the industry and consumers alike are left with exactly what NVIDIA has chosen to show — no more, no less.
