SignalTech/AI

Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s

Running 125B Qwen model on RTX 4090 at 100T/s signals major consumer hardware AI efficiency.

Context

What happened: A project called Strata released version 0.2.9, a free open source app for running the DeepSeek family model locally on desktop hardware, with installers for Windows, Linux, and macOS. Key numbers: ISTA-DAS sets range from 191 GB to 274 GB download, needing about 197 GB to 280 GB free disk. Why it matters: The page positions local chat, code, and image work without an account, and lists support for NVIDIA RTX 20 through 50 series, AMD Radeon, Intel Arc, Apple Silicon, and Strix Halo hardware.

What changed

A project called Strata released version 0.2.9, a free open source app for running the DeepSeek family model locally on desktop hardware, with installers for Windows, Linux, and macOS.

Why it matters

The page positions local chat, code, and image work without an account, and lists support for NVIDIA RTX 20 through 50 series, AMD Radeon, Intel Arc, Apple Silicon, and Strix Halo hardware.

Sources