DeepSeek671B~406 GB VRAM at Q4_K_M

DeepSeek R1 (671B) — VRAM, speed & local setup

Written by Jakub Rusinowski · Last updated January 20, 2025

The full massive MoE model. Requires enterprise hardware (H100/A100 clusters). State of the art.

DeepSeek R1 (671B) needs about 406 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.

Logic99

Creative90

Coding99

VRAM and speed by quantization

Quoted against NVIDIA RTX 4090 (24 GB). Includes the KV cache at 8K context, so it reads higher than the headline figure.

Quant	VRAM	Speed (est.)	Fit
Q2_K 2.63 bpw	222 GB	—	Too big
Q3_K_M 3.41 bpw	287.4 GB	—	Too big
Q4_K_M 4.83 bpw	406.5 GB	—	Too big
Q5_K_M 5.67 bpw	476.9 GB	—	Too big
Q6_K 6.56 bpw	551.6 GB	—	Too big
Q8_0 8.50 bpw	714.3 GB	—	Too big
F16 16.00 bpw	1343.4 GB	—	Too big

Black marker = usable memory on the NVIDIA RTX 4090 (24 GB). Estimates from the memory-bandwidth roofline on the methodology page. DeepSeek R1 (671B) VRAM calculator →

Get DeepSeek R1 (671B) running

DeepSeek R1 (671B) is not practical on a single consumer GPU: it needs about 405.9 GB at Q4_K_M, more than any single device in our catalogue holds with headroom. How multi-GPU inference works, or rent a multi-GPU server by the hour:

Deploy in the cloud now RunPod

or compare on Vast.ai

As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.

How to run DeepSeek R1 (671B)

Install Ollama, then run:

$ollama run deepseek-r1:671b

Weights on Hugging Face: deepseek-ai/DeepSeek-R1 ↗

Specifications

Curated — A hand-written entry from before this catalogue recorded its sources. The figures are long-standing but their provenance is not on file.

Parameters: 671 Billion
Context window: 128,000
Architecture: MoE
Provider: DeepSeek
Licence: MIT
Practical deployment: Multi-GPU server
Specified at: Q4_K_M
System RAM: 700 GB
Record updated: 2025-01-20

LicenceMITCommercial use permitted

Commercial use permitted. No usage restrictions beyond attribution.

Quality and use cases

Scores as published by the model’s authors or an independent evaluator — quality, not throughput, and not measured by us.

Best forresearchenterprise

Can I run DeepSeek R1 (671B) on my GPU?

Other DeepSeek R1 sizes

DeepSeek R1 (671B) — frequently asked questions

How much VRAM does DeepSeek R1 (671B) need?

About 406 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.

Does DeepSeek R1 (671B) run on an RTX 4090 (24 GB)?

No. DeepSeek R1 (671B) needs about 406 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.

How do I run DeepSeek R1 (671B) locally?

Install Ollama and run `ollama run deepseek-r1:671b`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

What other sizes does DeepSeek R1 come in?

DeepSeek R1 Distill Llama 8B (6 GB), DeepSeek R1 Distill Qwen 32B (20 GB), DeepSeek R1 Distill Qwen 14B (9 GB), DeepSeek R1 (671B) (406 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.