Breaking

Developer Claims Tool Runs 284B-Parameter DeepSeek Model on CPU and 64 GB RAM Without a GPU

November 9, 2026 at 02:46 EST

Open Source
Infra & Chips

A developer has announced llama-modchip, a tool that the developer says can run DeepSeek's 284-billion-parameter model entirely in a computer's system memory, without a GPU.

Developer announcement · llama-modchip

A 284B-parameter model on 64 GB of RAM, with no GPU: the claim

The developer says llama-modchip runs DeepSeek's 284-billion-parameter model entirely in system memory on a CPU alone. The claims are unverified, but if accurate they would bring a model this size within reach of consumer hardware.

Claimed figures at a glance

284B

parameters in the DeepSeek model

64 GB

system RAM the full model is claimed to fit in

9–15

words per second claimed throughput

16–32 GB

laptop RAM said to run 30B models

Parameter count, drawn to scale

Each block = 10B parameters. The 284B model is about 9.5× the 30B model.

30B

laptop-class model

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.