Strata 0.2.7 is an update that enhances the speed and compatibility of Qwen and DeepSeek on AMD and NVIDIA hardware, achieving significant performance gains, especially for 12 GB NVIDIA cards with improvements up to 37%. The update introduces support for Radeon RX 9000 and AI PRO R9700 cards, allows DeepSeek to run on Ryzen AI Max PCs, and features faster prompt reading and a redesigned web chat interface. Developed by Niko Veit in Slovenia, Strata allows the running of a 125-billion-parameter model on standard gaming PCs without the need for cloud services, ensuring user privacy. Although the update is free and shows impressive performance improvements, it requires substantial RAM and disk space, and users may experience some initial slowdowns during setup

Strata 0.2.7 Boosts Qwen and DeepSeek Speed on AMD

Strata 0.2.7 focuses on speed and compatibility, with a default change that speeds up 12 GB NVIDIA cards by up to 37 percent on code answers. It adds Radeon RX 9000 and AI PRO R9700 support for faster Qwen decoding, lets DeepSeek run on Ryzen AI Max (Strix Halo) Windows PCs, and makes DeepSeek read prompts much faster across most machines. New features include picture-reading turned on by default for DeepSeek, up to 1M-token context support, layer-major prefill for two NVIDIA cards, and a redesigned web chat with saved chats, search, branching, and LaTeX rendering, plus French as a new interface language.

Strata 0.2.7 Boosts Qwen and DeepSeek Speed on AMD @ Linux Compatible