Can DeepSeek R1 Distill Llama 70B run on the RTX 5090?
Yes. At 8K context the best fit is Q2_K, using about 27.8 GB of the 30.4 GB usable on the RTX 5090. Estimated speed ceiling: ~68 tokens/sec.
Better quality with offload: 4-bit (Q4_K_M)
Runs with offload 4-bit (Q4_K_M) at 8K context: 54 of 80 layers on the RTX 5090 (about 29.4 GB) and the other 26 in system RAM (about 13.9 GB). Needs a PC with at least 32 GB of system RAM. Estimated speed: 1.6–3.2 tokens/sec (slow but usable), versus a bandwidth ceiling of ~42 tok/s if it fit fully in VRAM.
llama-server -hf unsloth/DeepSeek-R1-Distill-Llama-70B-GGUF:Q4_K_M -c 8192 -ngl 54
Estimate, not a benchmark. The command downloads the GGUF from Hugging Face. The speed range spans dual-channel DDR4-3200 to DDR5-6000 system RAM; your CPU, RAM speed and llama.cpp version move it. If llama.cpp runs out of GPU memory, lower -ngl by 2-3; if VRAM is left over, raise it for more speed. Recent llama.cpp builds can also pick the split automatically. How we estimate offload.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) | 26.5 GB Fits | 27.8 GB Tight | 35.3 GB No | 65.3 GB No | 68 |
| 3-bit (Q3_K_M) | 34.0 GB No | 35.3 GB No | 42.8 GB No | 72.8 GB No | — |
| 4-bit (Q4_K_M) | 42.1 GB No | 43.3 GB No | 50.8 GB No | 80.8 GB No | — |
| 5-bit (Q5_K_M) | 48.7 GB No | 50.0 GB No | 57.5 GB No | 87.5 GB No | — |
| 6-bit (Q6_K) | 55.8 GB No | 57.0 GB No | 64.5 GB No | 94.5 GB No | — |
| 8-bit (Q8_0) | 71.7 GB No | 73.0 GB No | 80.5 GB No | 110 GB No | — |
| 16-bit (F16/BF16) | 133 GB No | 135 GB No | 142 GB No | 172 GB No | — |
• real GGUF file size. Speed is a bandwidth ceiling (1792 GB/s), not a benchmark. Methodology.
What fits instead
- Cheapest listed GPU that runs DeepSeek R1 Distill Llama 70B: 2x RTX 5090 (2× 32 GB, launch MSRP $3,998) at Q5_K_M, fully in VRAM at 8K context.
- Largest popular model that fits the RTX 5090: Qwen3.6 35B A3B (36B, MoE) at Q5_K_M, about 26.3 GB at 8K context.
"Runs" and "fits" mean the whole model in GPU memory at 8K context (4-bit or better for the model pick). Prices are US launch MSRPs, not today's street prices. "Popular" means the 60 most-downloaded models we track.
More
DeepSeek R1 Distill Llama 70B on every GPU and Mac · Every model that runs on the RTX 5090 · Other models that fit 32 GB VRAM · Try other settings in the calculator
Guides: Best GPU for local LLMs by budget · Can I run gpt-oss, Llama 70B or Qwen3? · Mac vs PC for local LLMs
Data: Hugging Face Hub (deepseek-ai/DeepSeek-R1-Distill-Llama-70B), NVIDIA specs. Model data last checked 2026-11-17; refreshed hourly.