The Lab

QuietWatts is one person, a power meter, and the machines on the bench. Every machine here was bought with my own money. The site exists to answer home-server and local-AI questions with measurements instead of guesses.

The hardware

• NVIDIA DGX Spark, GPU inference. Specification from NVIDIA's DGX Spark hardware page. Inference numbers live on /benchmarks; nothing here is a measured tokens/sec figure.
• Minisforum N5 Pro, storage and CPU inference. Full specification goes up alongside the first run.

Spark specification on this page is from NVIDIA, not from a tokens/sec figure. Published inference runs live on the benchmark database, measured under the published method. The N5 Pro's full card still waits on a run of that machine.

The rules

• A number gets published only if it came off this bench under the published method, or carries a citation to its source.
• Publication dates are set once. An updated date moves only when the content actually changes.
• Affiliate payout never touches a ranking. When the worse-paying product is better, it wins.
• Gear is only described as tested after it has been on this bench. Research-based picks are labeled as exactly that.

Why this site exists

Most of what ranks for home-server and local-AI searches was written without touching the hardware. The fix is boring and slow: buy the gear, measure it, publish the method next to the result, and re-run when the software moves. That is the whole plan.

Corrections and review-unit enquiries go through the contact page.