you ‧ notes ‧ unread ‧ starred ‧ add url

@basecampbernie: $300 mini PC running 26B parameter AI models at 20 tok/s. Minisforum UM790 Pro (

https://x.com/basecampbernie/status/2040326984446935059

april 2026 ‧ twitter twitter-post imported ‧ open in karakeep

by Base Camp Bernie ‧ X (formerly Twitter) ‧ published 2026.04.04 ‧ saved copy from 2026.08.28

From X bookmark — @basecampbernie (Base Camp Bernie), Apr 04 2026 https://x.com/basecampbernie/status/2040326984446935059

reader ‧ summary ‧ screenshot

open reader view in a new tab

A $300 mini PC, the Minisforum UM790 Pro with an AMD Radeon 780M iGPU and 48GB DDR5 RAM, can run 26 billion parameter AI models at around 20 tokens per second by using unified memory shared between the GPU and system RAM. Despite the GPU reporting only 4GB VRAM, it accesses over 21GB of system RAM at high bandwidth, allowing efficient AI workloads like Gemma 4 and Qwen3.5. This setup supports advanced local AI workflows with no need for API keys and offers a powerful, compact, and cost-effective solution, though DDR5 RAM pricing remains a limiting factor.

regenerate summary

page description:

$300 mini PC running 26B parameter AI models at 20 tok/s. Minisforum UM790 Pro ($351) + AMD Radeon 780M iGPU + 48GB DDR5-5600 + 1TB NVMe. The secret: the 780M has no dedicated VRAM. It shares your DDR5 via unified memory. The BIOS says “4GB VRAM” but Vulkan sees the full pool.

« back to bookmarks