Asus ProArt P16 evaluate: This 128GB RTX Spark laptop computer beat my RTX 5090
The ProArt completed nearly 20 minutes earlier, making it roughly 1.6 occasions sooner at finishing this specific monster workload. The 5090 run generated considerably extra completion tokens, 92,270 versus 79,903, so don’t learn this as a laboratory-perfect tokens-per-second comparability, and positively don’t learn it as “RTX Spark is quicker than the RTX 5090 at AI.” It isn’t. Give the 5090 a workload that matches comfortably inside its 32GB of VRAM and its vastly larger horsepower can run wild.
However gpt-oss-120b doesn’t match comfortably inside 32GB. My desktop can lean on one other 96GB of standard system RAM, however the mannequin is working throughout separate swimming pools of reminiscence. RTX Spark’s Grace CPU and Blackwell GPU share one gigantic pool, and I’d configured the ProArt to make roughly 96GB out there to the GPU, although my check crossed over into system reminiscence there, too. For this outsized workload, that architectural distinction mattered sufficient for the a lot slower laptop computer to complete first. By a lot.
The ProArt paid for it in battery life. I began the run at one hundred pc; 33 minutes later it sat at 54 p.c. However holy shit, that’s attention-grabbing. Nvidia caught 128GB of unified reminiscence on this factor so it may sort out monumental AI fashions, and I’d discovered an actual workload that genuinely put it to make use of. This 3.9-pound laptop computer completed the job nearly 20 minutes sooner than the monstrous RTX 5090 desktop sitting beside it.
The $7,000 query
After all, native AI isn’t free. This laptop computer prices 7,000 freaking {dollars} earlier than we even begin speaking about electrical energy, depreciation, or the frankly silly variety of hours I’ve spent screwing round with fashions and prompts. However the scale of my workload had gotten fairly silly, too.
