Long Term Review: Asus ProArt P16 RTX Spark – AI Agent-ing On Spark

In the latest installment of the in-depth analysis of the NVIDIA RTX Spark laptop, specifically the Asus ProArt P16 RTX Spark, the focus shifts to the AI capabilities that the device is known for. As previously discussed in the first part of the review, which detailed the hardware and specifications of both the laptop and the chipset, this particular RTX Spark model is equipped with a 20-core version featuring 6,144 Blackwell CUDA cores.

However, it's worth noting that the laptop only has half the memory of its counterpart, with just 64GB of LPDDR5-9600 Unified Memory.

NVIDIA has offered various methods to assess the RTX Spark's AI prowess, both when the device is online and when it's offline. This testing is accomplished through the use of Hermes and Unsloth agents. Hermes has been the go-to agent for the reviewer's preliminary attempts at coding and tinkering, while the latter has been utilized for more extensive tasks. NVIDIA provided guidance on how to utilize these agents for testing the RTX Spark laptop and its AI features.

One of the first tests undertaken was the creation of a simple dashboard using Hermes to monitor CPU, GPU, and logical processor activities. The process took roughly eight minutes on average over two runs, with the first run taking the longest at ten minutes. Another agent used was Unsloth, with Qwen3.8 27b serving as the language model.

This testing took about eight minutes on average, both online and offline. When using Qwen3.5-9b, the online training took around six minutes, while the offline training took an average of five minutes.

However, some benchmarks like MLPerf Client 2.0.1 encountered initial issues and failed to run the benchmark tests due to various reasons. Restarting the laptop resolved this issue, and Qwen3.8 27b and Flux 2 Klein 4b were able to complete their runs in an average of eight minutes. Another program, SpeedBench, which measures token-generation speeds with realistic prompts, also failed to launch at first. Subsequent attempts, however, succeeded in completing the runs in an average of eight minutes.

In terms of practicality, the reviewer doesn't foresee themselves relying heavily on AI agents for day-to-day operations. They see themselves using AI primarily for tasks such as spellchecking, rephrasing phrases and sentences, and building simple dashboards to monitor the laptop's performance. The AI agents can be beneficial for testing, exploring functionalities, and having fun, but they don't expect to rely on them extensively.

Despite this, the reviewer acknowledges the impressive capabilities of the RTX Spark laptop in handling AI tasks, whether online or offline.