HP’s first RTX Spark laptop is, by most measures, a fantastic machine. After two weeks with the HP OmniBook Ultra 16, reviewers found it games without issues and packs one of the best keyboards and OLED screens they have used. But there is one catch: if you want it for the local AI work NVIDIA is pitching hardest, the base 32GB model hits a wall.
HP OmniBook Ultra 16 (Credits: How-To Geek)
HP OmniBook Ultra 16 Review: Great Laptop, Tight Memory for Local AI
The OmniBook Ultra 16 is built around NVIDIA’s RTX Spark N1X chip, pairing a 20-core Grace Arm CPU with a Blackwell RTX GPU packing 6,144 CUDA cores. It sports a 16-inch 2880×1800 120Hz OLED display rated at up to 1,100 nits, a backlit latticeless keyboard with a haptic touchpad, and HP rates its battery at up to 17 hours. Multi-core CPU performance led every laptop in the review’s comparison group.
The base model starts at $3,200 for 32GB of RAM and 1TB of storage, available from Best Buy or HP, with upgraded configurations offering 64GB of RAM and 2TB of storage or 128GB of RAM and 2TB of storage directly from HP.
Why 32GB Is Not Enough for Agent-Class Local AI
The problem is how that 32GB gets used. By default, the memory is split into two chunks: 12GB dedicated to the Blackwell GPU and 10GB shared between the system and GPU, leaving a total budget of around 22GB once Windows reserves its share. In practice, most AI apps saw only around 20GB of VRAM available.
That proved too little for the kind of local AI agents NVIDIA is promoting. Trying to run Hermes, an agentic tool, the reviewer found that no model in the Qwen 3.6-35B-A3B family above a heavily compressed 2-bit version would load. Switching to LM Studio, they managed to load a 4-bit version as a partial offload, but only after a fight to get it running reliably at the 64K context Hermes needs.
Setting up local AI also took manual tuning, and the review notes that app defaults can mislead users about what will actually run.
Should You Buy the OmniBook Ultra 16?
For gaming, general productivity and everyday use, the OmniBook Ultra 16 looks like an excellent laptop: fast, beautifully screened and superbly built. But buyers whose main goal is running agent-class local AI models should think carefully before choosing the 32GB configuration. The review’s clear takeaway is that the higher-memory configurations exist for a reason, and for local AI work, they are the ones to consider.
In short: great RTX Spark laptop, but don’t buy the base model for local AI and expect it to fly.
Also Read: Don’t Throw Away Your Old Laptop: Turn It Into a Home Server and Backup Machine for Your New One
Source: How-To Geek — read the full story there.