# Sam Witteveen — '3,000 Tokens/Sec Locally: Testing the ZGX Fury'

> Sam Witteveen benchmarks HP's ZGX Fury, a compact GB300 machine a whole team can share for local AI.

- Org: Sam Witteveen
- Date: 2026-10-30
- Category: video
- Source: https://www.youtube.com/watch?v=lEFFa0Z5BIE
- Canonical page: https://ai-tldr.dev/releases/sam-witteveen-zgx-fury-oct30/

Sam Witteveen tests HP's ZGX Fury, a shared desktop AI machine built on NVIDIA's GB300 chip with 748 GB of memory, and shows local inference speeds, software and use cases.

## What it is

'3,000 Tokens/Sec Locally: Testing the ZGX Fury' went up on Sam Witteveen's channel on 30 October 2026. The video reviews HP's ZGX Fury, a larger sibling of the ZGX Nano (DGX Spark) built on NVIDIA's GB300 chipset with 748 GB of memory.

## How it works

The video explains how the machine's memory works, then covers HP's ZGX Runtime and Z Toolkit software and its Red Hat certification. It follows with benchmarks and use cases such as running agentic systems and on-prem AI.

## Why it matters

Teams that cannot send data to a cloud API need local hardware big enough for large models. A shared machine with this much memory is one option, and the video shows what speed it reaches.

## Who it's for

teams planning on-prem AI

## Links

- [Watch on YouTube](https://www.youtube.com/watch?v=lEFFa0Z5BIE)
- [Sam Witteveen](https://www.youtube.com/channel/UC55ODQSvARtgSyc8ThfiepQ)