EXPERIMENT, NOT A LAUNCH: a hands-on test by XDA Developers, published on October 25, 2026, had an HP OmniBook Ultra 16 run a local 125-billion-parameter language model and a Minecraft session at around 120 frames per second at the same time.
The laptop used in the test combines the NVIDIA RTX Spark platform with a 20-core Arm CPU, a 6,144-core Blackwell GPU and 128 GB of unified memory. The chosen model was Qwen3.8 Flash Next, a 125-billion-parameter mixture-of-experts model, served locally by llama.cpp from an NVIDIA recipe for RTX Spark.
The Minecraft bot was driven by the open Mindcraft project, built on Mineflayer: the model decides what to do and narrates its actions in the game chat, while Mindcraft handles movement, mining and inventory. In practice, the author asked for ten birch logs and the bot walked to the trees and started chopping; later, it dropped eight saplings at the player's feet. Measured inference was about 46 tokens per second at short context and just under 42 tokens per second at 32,000 tokens.
There are important nuances. Only about 6 billion parameters are active per token, and a 51-billion-parameter n-gram embedding is streamed from storage instead of sitting entirely in memory. The bot also makes plenty of mistakes: it is slow to respond, promises items it does not have, jumps into holes and disconnects when Mineflayer's unstuck system gives up after 20 seconds without moving. The author notes that these navigation stumbles come from Mineflayer's pathfinding, not from the model, and that Minecraft is not a demanding game: a heavy title would suffer far more with a model of this size running alongside it. In addition, NVIDIA's NVIGI SDK feature CUDA in Graphics is not yet supported on RTX Spark, so the two processes shared the GPU without that optimization.
Context helps to read the test: HP introduced the OmniBook Ultra 16 and OmniBook X 14 with RTX Spark in June 2026, as the world's thinnest RTX Spark laptops, with up to 128 GB of unified memory. The experiment does not prove that any heavy game can run next to a giant model, but it shows that local AI on a thin laptop is no longer just press-release promise: it already shares the machine with a real game, with visible flaws and measurable performance.
Sources: XDA Developers — I played Minecraft with a 125B local LLM on an RTX Spark laptop, and the game still held 120 FPS; WiseVoter — Laptop Runs Massive AI Model and Minecraft Simultaneously; HP (via WebWire) — HP Debuts PCs Built for the Next Wave of Windows PC Experiences Powered by NVIDIA RTX Spark.
Image credit: HP press-release image for PCs with NVIDIA RTX Spark, reproduced by WebWire.
Transparency: this article was prepared with the assistance of artificial intelligence, based on cross-checking public social media posts and the internet sources listed above. Consult the original sources to confirm details before making any decision.
— Muse
