Llama-modchip developers claim DeepSeek model runs on a 64 GB CPU system

Published Sunday, November 8, 2026 · 8:08 AM EST

The developers of llama-modchip say their software runs the 284.33B-parameter DeepSeek-V4-Flash model on a CPU system with 64 GB of RAM, without a GPU. The project provides downloadable model files and a patched executable, and reports operation at a 256K context length, the amount of text the model can process at once. The developers attribute the reduced memory requirement to compression and changes to model computation. They list a 56.56 GiB checkpoint, compared with a 76.87 GiB baseline that exceeds a 64 GB Linux system's memory. The custom tensor format requires their patched software; unpatched llama.cpp, Ollama and LM Studio cannot open it.

Sources

GitHub

In this story

Meta DeepSeek

Published by Tech & Business, a media brand covering technology and business. This story was sourced from GitHub and reviewed by the T&B editorial agent team.

Back to Newswire

Keep reading

Full wire

Robotics Capital

GenRobot AI raises $100 million for robot training data infrastructure

Beijing-based GenRobot AI has completed a $100 million Series A+ funding round, TechNode reported. The company develops equipment and software that collect human demonstrations for use in training robots, providing data infrastructure for companies developing robots.

Infrastructure

Blizzard used 3,924 virtual machines for World of Warcraft: Forever launch

Blizzard used 3,924 virtual machines with 70,480 CPU cores and 251,136 GB of memory to support the launch of World of Warcraft: Forever, Tom's Hardware reported. Players and media praised it as one of the company's smoothest launches.

Infrastructure AI

Inner Mongolia AI campus built 120MW within a year

A wind-powered 800MW AI data center campus in Inner Mongolia built 120MW within a year, SemiAnalysis reported. Its account describes accelerating construction in China's fastest-growing AI data center region, citing low building costs and no public backlash.

Meta ships specialized models with Active Memory and Active Focus

Meta is shipping specialized versions of its existing models to all Muse and Meta AI users, The New Stack reported. The rollout brings Active Memory and Active Focus to those users.

AI Infrastructure

Vercel adds DeepSeek search and page retrieval to AI Gateway

Vercel announced that DeepSeek Search and DeepSeek Fetch are available through AI Gateway for any model that supports tool calling. Developers can use one Gateway API key without creating a separate DeepSeek account and retain the tools when switching model providers.

AI Capital

OpenAI closes roughly $11 billion in funding at $1.25 trillion valuation

OpenAI has received roughly $11 billion in new funding, the initial portion of a round valuing it at $1.25 trillion, Bloomberg reported, citing people familiar with the matter. The money had been wired by Friday.