# Sam Witteveen — 'ZGX Fury - How Fast Can it Go?'

> Sam Witteveen puts HP's ZGX Fury through speed tests and real local AI workloads.

- Org: Sam Witteveen
- Date: 2026-11-16
- Category: video
- Source: https://www.youtube.com/watch?v=JVu2KicZACg
- Canonical page: https://ai-tldr.dev/releases/sam-witteveen-zgx-fury-speed-nov16/

## In short

Sam Witteveen returns to HP's ZGX Fury to test which models it can run and how fast, from memory math and speculative decoding to offline coding agents, private RAG and video generation.

## What it is

'ZGX Fury - How Fast Can it Go?' went up on Sam Witteveen's channel on 16 November 2026. It is his second video on HP's ZGX Fury desktop AI machine, this time focused on what models fit, how fast they run and what people can use it for. The video is sponsored by HP and NVIDIA.

## How it works

The chapters start with memory math for whether a model fits, HBM versus unified memory, and prefill versus decode. Speculative decoding, concurrent users and dense versus mixture-of-experts models come next, followed by measured model speeds.

## Why it matters

Local inference speed decides whether agents are usable on private hardware, and the video argues that agents need faster tokens. The second half tests concrete jobs: offline coding agents, private local RAG, audio and video transcription, and video generation.

## Who it's for

teams planning on-prem AI

## Links

- [Watch on YouTube](https://www.youtube.com/watch?v=JVu2KicZACg)
- [Sam Witteveen](https://www.youtube.com/channel/UC55ODQSvARtgSyc8ThfiepQ)
