Loading page…
His initial review puts the high-memory ASUS machine in a different workload class from two 32GB systems. More model options do not mean an easier start.
Listen to this story
The review frames RTX Spark as a capacity-versus-usability tradeoff: one high-memory configuration may support unusually large local workloads, but converting that headroom into a working setup remains difficult. In his preliminary Oct. 16 assessment, Paul Thurrott says the 128GB ASUS ProArt P16 can run 120-billion-parameter models with up to one million tokens of context through local agents; those figures are not independently verified benchmarks. He also finds local AI less capable than frontier cloud services and harder to get started with.
Thurrott reviewed three N1X machines: Lenovo’s Pro 9n 15 has an 18-core CPU and 32GB of RAM, while the HP OmniBook Ultra 16 has 32GB.
Only the ASUS ProArt P16 review unit has 128GB of unified memory, which Thurrott associates with the most demanding local workloads.
Thurrott sees lower-memory RTX Spark configurations as better suited to creators, developers and traditional workstation use.
Paul Thurrott’s October 16 RTX Spark review adds a context-capacity claim to the local-AI laptop pitch. He says the 128GB ASUS ProArt P16 can run 120-billion-parameter models with up to one million tokens of context using local agents. His assessment is preliminary, and the capacity figures are not independently verified benchmark results.
Nvidia and Microsoft are promoting RTX Spark as a class of Windows PCs for local AI, creative work, development and gaming. Thurrott has three machines in hand, giving his initial assessment a range of configurations rather than a single high-end system.
The review units span two N1X processor configurations. Memory creates another clear division: the Lenovo and HP each have 32GB, while the ASUS has the full 128GB supported by RTX Spark.
His provisional view places the 128GB configuration with AI enthusiasts and professionals pursuing the most demanding local workloads in Windows 11. He sees lower-memory configurations as a better fit for creators, developers and more traditional workstation users. That is a distinction between intended workloads, not a verdict that every RTX Spark machine serves the same audience.
And while RTX Spark does make a much broader range of local AI models available than is true of other PCs, that also amps up the complexity.
Paul Thurrott, Thurrott.com
Thurrott describes local AI as harder to get running than cloud services such as Copilot, Claude or ChatGPT, where users can open an app and begin prompting. In his assessment, local AI also trails frontier cloud AI functionally.
He finds it difficult to know where to start with the expanded model selection. His early assessment separates two questions: what the hardware can accommodate, and how readily a user can turn that capacity into a working local-AI setup.
Loading discussion...
Join the conversation
Explain when extra choice becomes worth the effort.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.