BBrainOutput
DeepSeek·Reasoning·MIT·DeepSeek·2025

DeepSeek-R1 671B (MoE): Hardware & Business Fit

  • Reasoning
  • Code
  • Long context

The full DeepSeek-R1, included to anchor the top of the reasoning tier. Only the distilled variants are realistic for single-box local deployment. Figures are placeholders.

Parameters
~671B (≈37B active, MoE)
Context
~128K tokens
Deployment
cloud
VRAM @ 4-bit
~400GB

What DeepSeek-R1 671B (MoE) is good for

  • Frontier reasoning, self-hosted at scale
  • Research-grade analysis
frontier reasoningMoE efficiencyMIT license

Best quantization choices

Approximate memory per quantization (weights + KV cache at modest context). Treat as ±.

Quant~MemoryWhen to use
Q4_K_M~400GBBest size/quality trade-off — the usual default for local serving.
Q8_0~700GBHigher fidelity; ~1.7× the memory of 4-bit.
FP16~1340GBFull precision; largest footprint, best quality.

Run DeepSeek-R1 671B (MoE) locally

Pull and run with Ollama, or grab the weights from Hugging Face.

$ ollama run deepseek-r1:671b
Hugging Face repo
deepseek-ai/DeepSeek-R1

Compatible hardware

Devices from our catalog graded for DeepSeek-R1 671B (MoE), best fit first.

  • Supermicro 8x H100 SuperServer
    Supermicro · AI Servers

    Fits at Q4_K_M (~400GB) with ~163.2GB headroom — about 1 concurrent instance.

    Q4_K_M · ~400GBRuns well
  • Dell PowerEdge XE9680
    Dell · AI Servers

    Fits at Q4_K_M (~400GB) with ~163.2GB headroom — about 1 concurrent instance.

    Q4_K_M · ~400GBRuns well
  • NVIDIA B200 (placeholder)
    NVIDIA · Datacenter GPUs

    Even the smallest quantization (~400GB) exceeds usable memory (~169GB). Choose a smaller model or step up the hardware.

    Not recommended
  • AMD Instinct MI300X
    AMD · Datacenter GPUs

    Even the smallest quantization (~400GB) exceeds usable memory (~169GB). Choose a smaller model or step up the hardware.

    Not recommended
  • Cloud B200 (Blackwell profile, to verify)
    Cloud · Cloud GPU Profiles

    Even the smallest quantization (~400GB) exceeds usable memory (~158.4GB). Choose a smaller model or step up the hardware.

    Not recommended
  • NVIDIA H200 (141GB)
    NVIDIA · Datacenter GPUs

    Even the smallest quantization (~400GB) exceeds usable memory (~124.1GB). Choose a smaller model or step up the hardware.

    Not recommended
  • Cloud H200 141GB (profile)
    Cloud · Cloud GPU Profiles

    Even the smallest quantization (~400GB) exceeds usable memory (~124.1GB). Choose a smaller model or step up the hardware.

    Not recommended
  • NVIDIA H100 (80GB)
    NVIDIA · Datacenter GPUs

    Even the smallest quantization (~400GB) exceeds usable memory (~70.4GB). Choose a smaller model or step up the hardware.

    Not recommended
  • Cloud H100 80GB (profile)
    Cloud · Cloud GPU Profiles

    Even the smallest quantization (~400GB) exceeds usable memory (~70.4GB). Choose a smaller model or step up the hardware.

    Not recommended
  • NVIDIA RTX PRO 6000 Blackwell
    NVIDIA · Professional GPUs

    Even the smallest quantization (~400GB) exceeds usable memory (~84.5GB). Choose a smaller model or step up the hardware.

    Not recommended

Use inside the AI Business OS

DeepSeek-R1 671B (MoE) suits these AI Business OS agent archetypes:

A model is only the engine. Inside the AI Business OS it is wrapped with permissions, tools, connectors, RAG and audit so it can actually do business work safely — see how the AI Business OS works →

Frequently asked questions

What hardware do I need to run DeepSeek-R1 671B (MoE)?+

At 4-bit you need roughly ~400GB of usable memory. The minimum self-hostable option in our catalog is the Supermicro 8x H100 SuperServer. For a comfortable run we recommend the Supermicro 8x H100 SuperServer.

Which quantization should I use for DeepSeek-R1 671B (MoE)?+

Q4_K_M is the usual default — the best size/quality trade-off. Step up to Q8_0 or FP16 if you have spare memory and want higher fidelity.

Should I run DeepSeek-R1 671B (MoE) locally or in the cloud?+

Cloud / API is recommended for DeepSeek-R1 671B (MoE). Its size or hosting model makes the cloud the practical choice; pair with smaller local models for everyday private work.

Other sizes in the DeepSeek family

All DeepSeek models →

Same family, different size. Pick the variant that fits your hardware.

Related models

Similar picks — family siblings and nearest-size models of the same kind.

Use DeepSeek-R1 671B (MoE) inside your AI Business OS

BrainOutput helps you run DeepSeek-R1 671B (MoE) as a private business agent — wrapped with the tools, connectors, RAG and guardrails it needs to do real work on hardware you control.

Use this model in your AI Business OS