Nvidia RTX Spark: The ARM Chip That Could Change Sim Racing PCs
At Computex 2026, Nvidia unveiled the RTX Spark (codename N1X) — a historic first for the company: a system-on-chip combining GPU and CPU in a single package, designed for compact Windows PCs. The stated target is on-device AI and content creation, but the announced specs — an RTX 5070-class GPU in a chassis the size of a shoebox — raise interesting questions for sim racers looking to shrink their setups.
A new architecture: Blackwell + ARM in one package
RTX Spark is built around a two-chiplet design connected by a NVLink-C2C silicon bridge at 600 GB/s bandwidth. The first chiplet is a Blackwell GPU with 6,144 CUDA cores — equivalent to an RTX 5070 mobile — with DLSS 4.5 and second-generation Ray Reconstruction support. The second is a 20-core ARM Grace CPU co-designed with MediaTek: 10 high-performance Cortex-X925 cores and 10 efficiency Cortex-A725 cores, clocked up to 4.1 GHz.
The whole thing is built on TSMC’s 3nm process with around 70 billion transistors, and shares up to 128 GB of unified LPDDR5X memory between CPU and GPU — an approach Apple popularised with its M-series chips. The first devices target very compact formats: the ASUS ProArt Mini PC measures just 150 × 150 × 51 mm, with Dell, Lenovo, and MSI also among launch partners.
What it could change for sim racing
For a sim racer, the primary appeal is form factor. A PC capable of running simulators at 1440p, 100+ fps, with ray tracing and DLSS 4 — in a chassis the size of a shoebox — is a concrete gain for anyone integrating their PC into a cockpit or behind a triple-screen setup where space is tight. The reduced power consumption of an SoC architecture would also limit heat output during long sessions.
The unified memory pool of up to 128 GB is another notable point. In a traditional build, VRAM is the bottleneck for triple screens or high-resolution VR. With a shared memory space of this size, that problem disappears in theory — all simulators, textures, and SimHub overlays fit in the same pool without CPU/GPU arbitration.
The open questions
The most significant concern is simulator compatibility. iRacing, Assetto Corsa, rFactor 2, Automobilista 2 — all are x86 applications compiled for Intel/AMD processors. On Windows ARM, they would run via emulation, much like x86 apps on Apple Silicon before native ports. Emulation works, but carries a performance penalty, particularly on single-threaded workloads where advanced physics simulators are most demanding.
Early benchmarks are mixed on this front: the ARM Grace CPU runs around 30% slower in single-thread compared to Apple M5 Max, and nearly 20% behind the Snapdragon X Elite. Racing simulators with complex physics engines — iRacing being the prime example — are notoriously single-thread dependent. The chip’s TDP has not yet been disclosed, making it difficult to assess its behaviour under sustained load.
RTX Spark is an announcement that deserves sim racers’ attention, but one that calls for caution before enthusiasm. A Blackwell GPU in such a compact form factor is genuinely new, and potentially very interesting for space-constrained setups. But until the major simulators are ported natively to ARM, x86 emulation will remain a limiting factor that is difficult to quantify. One to watch very closely when real-world benchmarks arrive with the first devices in late 2026.
Specifications
ℹ️ Information collected in June 2026 at Computex. No pricing has yet been announced by Nvidia.