453 points57 comments
Run 284B DeepSeek on a laptop with no GPU
What is Llama-modchip: 284B DeepSeek on 64GB RAM, no GPU?
llama-modchip lets you load the 284B DeepSeek-V4-Flash mixture-of-experts model entirely in system RAM, fitting into 64 GB without any GPU. It uses patched ggml/llama.cpp kernels and 1.71-bits-per-weight quantization to deliver about 44 tokens per second on a 64-GB desktop.
Who it is for
Developers who want to experiment with massive MoE models can prototype locally without spending on high-end GPUs.
The story so far
Early Signal first spotted Llama-modchip: 284B DeepSeek on 64GB RAM, no GPU on November 8, 2026, with 453 points. On its busiest day it gained 122 points. Early Signal files it under LLMs, Inference.
Questions about Llama-modchip: 284B DeepSeek on 64GB RAM, no GPU
- What do I need to run Llama-modchip: 284B DeepSeek on 64GB RAM, no GPU?
- Linux x86_64 CPU with AVX-512/AVX2 and 64 GB RAM