PCWorld reported that the ASUS ProArt P16 with RTX Spark finished one gpt-oss-120b workload in 33:13, while a desktop with a GeForce RTX 5090 took 52:51. The ProArt took 19:38 less elapsed time, but produced fewer completion tokens, so the runs do not establish an equal-output throughput advantage.
The reported times and output
Brad Chacos’s PCWorld comparison recorded 79,903 completion tokens from the ProArt and 92,270 from the RTX 5090 desktop, a difference of 12,367 tokens. The elapsed-time gap is about 37.1% of the desktop’s 52:51 run. That percentage describes time alone, not how quickly the systems generated an equal amount of output.
| System | Test configuration | Elapsed time | Completion tokens |
| ASUS ProArt P16 with RTX Spark | 20-core Grace CPU, integrated Blackwell graphics, 128GB unified LPDDR5X memory; roughly 96GB allocated to graphics | 33:13 | 79,903 |
| Desktop with GeForce RTX 5090 | Intel Core i7-14700K, 96GB system RAM, RTX 5090 with 32GB GDDR7 | 52:51 | 92,270 |
What the systems ran
Both systems ran gpt-oss-120b with the same prompts, inputs, stage boundaries and inference settings. The timed portion excluded live web access and state acquisition. PCWorld’s account concerns this workload, not a broad comparison across AI models or applications.
The ProArt’s 128GB of LPDDR5X memory is shared between its CPU and integrated GPU. The RTX 5090 desktop, by contrast, had 32GB of dedicated GDDR7 video memory, or VRAM. ASUS’s MyASUS utility allowed up to 96GB of the laptop’s shared pool to be allocated to graphics; Chacos used roughly that amount for the largest AI workloads. NeoTeo’s earlier coverage of the ProArt P16 H7607 discusses the laptop’s wider local-AI and creative-work profile.
Why the laptop finished sooner in this workload
PCWorld attributed the ProArt’s shorter run to the model’s memory demands. The reviewer’s explanation was that the workload strained the RTX 5090’s 32GB of VRAM, while the ProArt could draw on a much larger shared memory pool. That is a workload-specific explanation, not a general rule that RTX Spark is faster for AI. Chacos noted that the RTX 5090 has substantially greater raw GPU horsepower and can perform strongly when a workload fits comfortably in its VRAM.
Battery use and sustained runs
The ProArt’s battery reading fell from 100% to 54% during the 33:13 comparison run, a drop of 46 percentage points. In a separate seven-part workflow, PCWorld reported 150,995 generated tokens over 415 seconds, while the battery reading went from 74% to 65%. The reviewer described that output as useful, though below the strongest frontier models.
PCWorld also reported loud fans, failed or truncated runs, context problems and adjustments to prompts and configurations during its local-AI work. Those practical hurdles accompanied the review’s experiments with large models; the reported 33:13 result alone does not describe every run.
The dated US price
In its October 29, 2026 report, PCWorld listed the reviewed 128GB ASUS ProArt P16 RTX Spark configuration at $6,999 in the United States.
