Skip to content
Run LLM Locally

DeepSeek R1 Distill Llama 8B on the RTX 3090

With 24 GB of VRAM, the RTX 3090 runs DeepSeek R1 Distill Llama 8B at Q4_K_M entirely on the GPU with an 8K context (about 6.7 GB needed). At Q8_0 it fits entirely.

ShareFacebookXWhatsAppLinkedInEmail

DeepSeek R1 Distill Llama 8B

Q4_K_M · 8K context

Excellent

The model needs about 6.7 GB and fits in the 23.3 GB available, with 16.6 GB to spare.

Your hardware
RTX 3090 24 GB · 64 GB RAM
Estimated memory
~6.7 GB estimate
VRAM (usable memory)
23.3 GB
GPU only
Yes
CPU/GPU offload
Not required
Generation speed
106–153 tok/s estimate
Largest context in fast memory
128K tokens
Confidence: low

Based on: memory bandwidth (936 GB/s) and model size

Memory breakdown

~6.7 GB (6.2 GB–7.1 GB) needed · estimate

Fits entirely in VRAM · 16.6 GB left

  • Model weights4.6 GB
  • KV cache1 GB
  • Runtime overhead0.8 GB
  • Safety margin0.3 GB
Assumptions: context 8K tokens · KV cache f16 · llama.cpp (CUDA) · flash attention on · weights = real file size

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

What makes this estimate less reliable

  • no benchmark for this hardware and model yet

Other quantizations on the RTX 3090

DeepSeek R1 Distill Llama 8B on RTX 3090 + 64 GB RAM · 8K context
DeepSeek R1 Distill Llama 8B Q8_08K~10.2 GB · 69–96 tok/s estimateExcellent
DeepSeek R1 Distill Llama 8B Q6_K8K~8.3 GB · 85–120 tok/s estimateExcellent
DeepSeek R1 Distill Llama 8B Q5_K_M8K~7.5 GB · 94–135 tok/s estimateExcellent
DeepSeek R1 Distill Llama 8B Q4_K_M8K~6.7 GB · 106–153 tok/s estimateExcellent
DeepSeek R1 Distill Llama 8B IQ4_XS8K~6.2 GB · 114–166 tok/s estimateExcellent
DeepSeek R1 Distill Llama 8B Q3_K_M8K~5.8 GB · 122–180 tok/s estimateExcellent

Alternatives

GPUs that run DeepSeek R1 Distill Llama 8B Q4_K_M entirely in VRAM

Other models that fit the RTX 3090

Where to buy a RTX 3090

Prices and links follow the market you choose.

GeForce RTX 3090

24 GB GDDR6X · 936 GB/s · 350 W

ExcellentRuns DeepSeek R1 Distill Llama 8B 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 3090 run DeepSeek R1 Distill Llama 8B?
The model needs about 6.7 GB and fits in the 23.3 GB available, with 16.6 GB to spare.
Which quantization should I use on the RTX 3090?
Q8_0 is the highest-precision quantization that fits entirely in VRAM at 8K context. ~10.2 GB / 23.3 GB.
What is the largest context on the RTX 3090?
At Q4_K_M, about 128K tokens fit entirely in VRAM with an f16 KV cache. A q8_0 KV cache roughly doubles that.
How fast is DeepSeek R1 Distill Llama 8B (RTX 3090)?
We estimate 106–153 tok/s (estimate, confidence low). Based on: memory bandwidth (936 GB/s) and model size.