In results PCWorld published November 4, 2026, the ASUS ProArt P16 completed a 120-billion-parameter gpt-oss task in 33 minutes 13 seconds, while a desktop with a GeForce RTX 5090 took 52 minutes 51 seconds. PCWorld attributed the difference in that run to the laptop’s larger pool of GPU-accessible memory, not to a faster GPU.

NeoTeo’s earlier RTX Spark benchmark report covered results from the Surface Laptop Ultra. This article looks at PCWorld’s ProArt P16 AI result and separate RTX Spark graphics and gaming tests.

Why the ProArt P16 finished the AI run sooner

PCWorld reported that both systems processed 472,084 tokens of OpenAI’s gpt-oss model. Its tested ProArt P16 had 128GB of unified memory, with about 96GB available to the GPU. The desktop had 32GB of GDDR7 video memory. PCWorld said the larger GPU-accessible memory pool helped the laptop complete this particular run sooner; the result does not mean its GPU is faster than the RTX 5090.

The elapsed-time gap was 19 minutes 38 seconds. That comparison applies to this model and run, not to local AI workloads in general.

Geekbench OpenCL compares one specific configuration

ETA PRIME’s review shows ASUS ProArt P16 benchmark runs and game tests, including its Forza Horizon 6 test through Prism.

ETA PRIME reported 117,129 points for the RTX Spark N1X in Geekbench OpenCL, a synthetic graphics-compute test. Its comparison gave the GeForce RTX 5070 Laptop GPU 113,759 points and the AMD Ryzen AI Max+ 395 graphics 86,218. Those scores describe this benchmark, not overall gaming performance.

ETA PRIME described its tested ASUS ProArt P16 as having a 20-core Arm CPU, a Blackwell GPU with 6,144 CUDA cores and 128GB of LPDDR5X unified memory running at 9,600 MT/s. In manual mode, the reviewer said the system sustained up to 85 watts shared between the CPU and GPU. These details apply to the tested configuration.

TestSystem and resultCondition
Local AI, gpt-oss with 120 billion parametersASUS ProArt P16: 472,084 tokens in 33:13; GeForce RTX 5090 desktop: 472,084 tokens in 52:51PCWorld’s reported run; it attributed the timing difference to GPU-accessible memory
Geekbench OpenCLRTX Spark N1X: 117,129 points; GeForce RTX 5070 Laptop GPU: 113,759; AMD Ryzen AI Max+ 395 graphics: 86,218ETA PRIME’s synthetic graphics-compute comparison
3DMark Port RoyalRTX Spark N1X: 9,212 pointsETA PRIME’s ray-tracing test; no comparison score is included here

ETA PRIME also reported the Port Royal result, but a score without a comparable result does not establish a ranking. Its ProArt P16 benchmark and game tests include the synthetic tests and the game-specific compatibility observation discussed next.

Gaming results include a Prism problem

PCWorld said the RTX Spark laptops it tested roughly kept pace with GeForce RTX 5070 discrete graphics in games. It also found that extra unified memory did not improve the games it tested, and that conventional GeForce laptops offered better gaming performance for less money.

ETA PRIME reported severe stuttering when running Forza Horizon 6 through Microsoft Prism, a software translation layer, and described the game as unplayable in that test. PCWorld separately said its Surface Laptop Ultra testing required a beta version of Adobe Premiere for proper GPU support. Each is a specific software observation, not a result for every game or application.

Battery observations depend on the laptop and workload

PCWorld reported about 10 hours of office-work battery life on its Microsoft Surface Laptop Ultra test. In a separate test, an AI run on the ProArt P16 used 46% of a full charge in 33 minutes. The first figure is an office-use result from one laptop; the second is a single AI workload on another.