At Computex 2026 NVIDIA unveiled RTX Spark, a Windows‑on‑Arm platform that promises the same RTX and AI capabilities traditionally found in its data‑center products. The first retail device to carry the new architecture is the ASUS ProArt P16 laptop, which arrives with the full‑die configuration of NVIDIA’s N1X system‑on‑chip.
Hardware Overview
The ProArt P16 is powered by the N1X SoC fabricated on TSMC’s 3 nm process. The chip integrates 20 CPU cores—ten Arm Cortex‑X925 performance cores that boost to 4.0‑4.1 GHz and ten Cortex‑A725 efficiency cores capped at 2.8 GHz—delivering 20 threads. Cache is split between 24 MB of L3, 25 MB of L2 and a total of 49 MB on‑chip memory. The integrated graphics engine is NVIDIA’s GB20B Blackwell GPU with 6144 CUDA cores across 48 streaming multiprocessors, 5th‑generation Tensor cores, RTX ray‑tracing units and support for DLSS, Reflex, ACE, G‑Sync, CUDA and NVFP4. The GPU can produce up to 31 TFLOPs of FP32 compute and 1000 TOPS of FP4 AI throughput, backed by an additional 24 MB of L2 cache.
Memory is supplied via a 256‑bit LPDDR5X interface capable of 9400 MT/s, yielding roughly 300 GB/s of bandwidth and a unified pool of up to 128 GB. The system fabric uses the CHI‑E coherency protocol and provides the GPU with up to 600 GB/s of aggregate bandwidth over a C2X link. Storage is a 2 TB PCIe Gen4 SSD in one of two M.2 slots, with the chassis supporting up to 16 TB total capacity.
Connectivity includes PCIe Gen5, four USB4 ports (40 Gbps each), DisplayPort 2.0, HDMI 2.1a, Wi‑Fi 7 and a suite of security processors (Dual Secure Root, SROOT, OSROOT, fTPM and discrete TPM). The laptop measures 1.29‑1.59 cm in thickness, weighs 1.77 kg, and houses a 99.9 Wh 4‑cell battery paired with a 140 W USB‑C charger. Two display options are offered: a 16‑inch 3840×2400 OLED panel at 700 nits (1600 HDR) with 120 Hz refresh, and a 2880×1800 OLED panel at 500 nits (1100 HDR) also up to 120 Hz.
Benchmark Results
Synthetic CPU testing shows mixed results. In 3DMark’s CPU Profile the RTX Spark configuration trails x86 competitors in both single‑core and multi‑threaded workloads, despite its higher core count. Memory bandwidth, however, is the fastest among the group thanks to the 256‑bit LPDDR5X subsystem. Blender rendering beats Intel’s Panther Lake but falls short of AMD’s Ryzen AI MAX+ 300 SoC. CPU‑Z reports the highest single‑ and multi‑core scores of the tested devices, while Cinebench places the N1X ahead of Ryzen AI MAX+ 395, Snapdragon X2 Elite and Intel Panther Lake in both single‑ and multi‑core tests. Geekbench 6 also ranks RTX Spark above Intel Panther Lake and AMD Ryzen AI MAX+ 300, though Procyon Office shows Panther Lake and Ryzen AI MAX leading.
GPU synthetic benchmarks favor the Blackwell GPU. In 3DMark Speed Way (ray‑tracing), Steel Nomad, Port Royal and Time Spy the RTX Spark device posts the top scores, outpacing Radeon 8060S by 66 % in Port Royal and delivering more than double the performance of the Arc B390. Fire Strike shows the RTX Spark nearly matching the Ryzen AI MAX+ 395, with graphics strength offsetting a modest CPU lag. Night Raid places the device ahead of the B390 and close to the Radeon 8060S.
AI and Gaming Performance
AI workloads demonstrate the platform’s strengths. UL Procyon benchmarks register the fastest results among the compared SoCs. In LM Studio, a range of large language models run comfortably, and newer community‑optimized models such as Nemotron Bolt and VL are noted to target 24 GB RTX Spark configurations. MLPerf Client 2.0.1 reports a GenAI throughput of 15.67 images per minute for the Flux 2 Klein 4‑billion‑parameter model, while an Agent test using Llama 3.1 8B Instruct with NVFP4 completes tool operations in roughly seven seconds.
Gaming tests focus on Cyberpunk 2077 at a 1200p medium preset using DLSS 4.5 and Frame‑Gen. The RTX Spark chip trails the Radeon 8060S when DLSS alone is applied, but Frame‑Gen pushes the laptop ahead of the Intel 395 configuration. Forza Horizon 6 testing is mentioned but details are incomplete in the source.
Overall, the ASUS ProArt P16 showcases the first consumer‑grade implementation of NVIDIA’s RTX Spark platform. Its combination of a high‑core‑count Arm CPU, a Blackwell‑based GPU and a large unified memory pool positions it as a strong contender for creators and AI developers seeking laptop‑based alternatives to cloud services.
Steve Lopez is a Senior Editorial Columnist and Health & Public Policy reporter for News Raise. Steve focuses on healthcare advancements, medical technologies, and public health policies.
