Skip to content
Run LLM Locally

DeepSeek R1 (671B) on the RTX 5090

DeepSeek R1 (671B) at Q4_K_M needs about 402 GB: the RTX 5090 (32 GB) cannot run it well, even with system RAM. At Q3_K_M it does not fit in VRAM.

ShareFacebookXWhatsAppLinkedInEmail

DeepSeek R1 (671B)

Q4_K_M · 8K context

Incompatible

The model needs about 402 GB; VRAM plus usable RAM give 82.2 GB. 319 GB are missing.

Your hardware
RTX 5090 32 GB · 64 GB RAM
Estimated memory
~402 GB estimate
VRAM (usable memory)
31 GB
GPU only
No
CPU/GPU offload
Required (51.2 GB)
Generation speed
Not available
Largest context in fast memory
None
Confidence: high

Memory breakdown

~402 GB (393 GB–410 GB) needed · estimate

Does not fit · 319 GB missing

  • Model weights381 GB
  • KV cache0.5 GB
  • Runtime overhead0.8 GB
  • Safety margin19.1 GB
Assumptions: context 8K tokens · KV cache f16 · llama.cpp (CUDA) · flash attention on · weights = real file size · compressed MLA KV cache

EstimateData updated 24 days agoSource: Run LLM Locally engine, Hugging Face config.json + model card

What makes this estimate less reliable

  • split CPU/GPU runs vary a lot with the CPU and PCIe link
  • system RAM speed unknown (dual-channel DDR4-3200 assumed)
  • mixture-of-experts: speed depends on expert routing

Other quantizations on the RTX 5090

DeepSeek R1 (671B) on RTX 5090 + 64 GB RAM · 8K context
DeepSeek R1 (671B) Q8_08K~699 GB · No dataIncompatible
DeepSeek R1 (671B) Q6_K8K~542 GB · No dataIncompatible
DeepSeek R1 (671B) Q5_K_M8K~469 GB · No dataIncompatible
DeepSeek R1 (671B) Q4_K_M8K~402 GB · No dataIncompatible
DeepSeek R1 (671B) IQ4_XS8K~353 GB · No dataIncompatible
DeepSeek R1 (671B) Q3_K_M8K~303 GB · No dataIncompatible

Alternatives

GPUs that run DeepSeek R1 (671B) Q4_K_M entirely in VRAM

    Other models that fit the RTX 5090

    Hardware that runs DeepSeek R1 (671B) Q4_K_M entirely in memory

    Prices and links follow the market you choose.

    Mac Studio M3 Ultra

    Apple Silicon

    Compatible with limitsRuns DeepSeek R1 (671B) Q4_K_M entirely in memory

    Price not available

    Availability unknown

    View on amazon.com

    Affiliate link

    No current offer imported for this market: the link opens a marketplace search.

    Frequently asked questions

    Can the RTX 5090 run DeepSeek R1 (671B)?
    The model needs about 402 GB; VRAM plus usable RAM give 82.2 GB. 319 GB are missing.
    Which quantization should I use on the RTX 5090?
    None fits entirely in VRAM at 8K context; the smallest, Q3_K_M, still needs offload.
    What is the largest context on the RTX 5090?
    At Q4_K_M, the model does not fit entirely in VRAM even at a short context.