VRAMNeeded Calculator

DeepSeek · VRAM requirements

DeepSeek-R1-0528-Qwen3-8B VRAM requirements

  • Parameters8.2B
  • Context128K
  • 4-bit file5.0 GB
  • Checked2026-10-28

Quick answer

At the popular 4-bit Q4_K_M quantization, DeepSeek-R1-0528-Qwen3-8B's weights are 5.0 GB, and an 8K-token context adds about 1.21 GB of KV cache, so plan for roughly 6.2 GB plus your engine's overhead.[4][3] That fits an 8 GB card such as the RTX 4060 with room for context. Unquantized BF16 weights are about 16.4 GB.[4]

DeepSeek-R1-0528-Qwen3-8B is from DeepSeek. Its Hugging Face repo was created on 2025-05-29, it is licensed MIT, and Hugging Face lists 8,190,735,360 parameters.[1] Native context is up to 131,072 tokens where the card documents it.[2]

Memory needed at each quantization

Quantization stores each weight in fewer bits. Smaller files fit on smaller GPUs, at some cost in quality. Below are GGUF builds from the bartowski (or listed) repo using exact Hugging Face file sizes (1 GB = 1,000,000,000 bytes), largest first.[4]

QuantizationWeights
BF16 (original)16.4 GB
Q8_08.7 GB
Q6_K_L7.0 GB
Q6_K6.7 GB
Q5_K_L6.2 GB
Q5_K_M5.9 GB
Q5_K_S5.7 GB
Q4_K_L5.5 GB
Q4_15.2 GB
Q4_K_M Common default5.0 GB
Q3_K_XL5.0 GB
Q4_K_S4.8 GB
IQ4_NL4.8 GB
Q4_04.8 GB
IQ4_XS4.6 GB
Q3_K_L4.4 GB
Q3_K_M4.1 GB
IQ3_M3.9 GB
Q2_K_L3.9 GB
Q3_K_S3.8 GB
IQ3_XS3.6 GB
IQ3_XXS3.4 GB
Q2_K3.3 GB
IQ2_M3.1 GB

Ollama's deepseek-r1:8b tag is the usual one-line install for the 8B R1 distill build when the library lists it.[5][6]

Context length adds memory (the KV cache)

36 layers × 8 KV heads × 128 head dim × 2 × 2 bytes. This is what Ollama's default deepseek-r1 tag (8b) runs.[3] An FP16 cache takes 147,456 bytes per token:

ContextKV cache (FP16)
4,096 tokens0.60 GB
8,192 tokens1.21 GB
32,768 tokens4.83 GB
131,072 tokens19.33 GB

Some engines can store the cache in 8-bit to roughly halve it.

Which GPUs and Macs it fits on

Each cell compares the weights plus an 8K-token KV cache with the device's memory.[7][8][9][10][11][12][13]

DeviceBF16 (original)17.6 GBQ8_09.9 GBQ6_K7.9 GBQ5_K_M7.1 GBQ4_K_M6.2 GBIQ4_XS5.8 GB
RTX 40608 GBNoNoTightFitsFitsFits
RTX 306012 GBNoFitsFitsFitsFitsFits
RTX 4060 Ti16 GBNoFitsFitsFitsFitsFits
RTX 309024 GBFitsFitsFitsFitsFitsFits
RTX 409024 GBFitsFitsFitsFitsFitsFits
RTX 509032 GBFitsFitsFitsFitsFitsFits
Mac16 GB unifiedNoFitsFitsFitsFitsFits
Mac24 GB unifiedFitsFitsFitsFitsFitsFits
Mac32 GB unifiedFitsFitsFitsFitsFitsFits

Column headers show weights + 8,192-token KV cache.

How to read it: "Fits" means weights + 8K cache use under 90% of memory. "Tight" means 90–100%. "No" means it won't fit on the GPU alone. The 90% line is our own rule of thumb, not a vendor figure. On a Mac, macOS also needs some of the unified memory, so treat Mac rows as optimistic.

How to run it

The easiest way is Ollama when the library carries this build: ollama run deepseek-r1:8b.[5] With llama.cpp or LM Studio, download a GGUF such as the Q4_K_M file from bartowski/deepseek-ai_DeepSeek-R1-0528-Qwen3-8B-GGUF.[4]

If the model is bigger than your VRAM, these tools can keep some layers in system RAM, but every token then waits on the slower memory, so expect much lower speed.

Cheapest GPU to rent for DeepSeek-R1-0528-Qwen3-8B

On-demand Secure Cloud prices listed on Runpod's pricing page (confirm before you rent — prices move):[14]

GPU (Runpod Secure Cloud (on-demand, as listed))VRAMPrice
RTX 409024 GB$0.89/hr
RTX 509032 GB$1.19/hr
RTX A600048 GB$0.59/hr
A100 PCIe80 GB$1.79/hr

Prices checked on 2026-10-28. They change often, so confirm on the provider's page. Rent on Runpod Referral link

Rent a GPU for this model
Affiliate partnerVast.aiConsumer GPU marketplace (4090/3090-class rentals)See GPUs →Referral linkRunpodPods and serverless GPUsSee GPUs →Works withE2E NetworksIndia GPU cloudSee GPUs →
Run it via API instead
Affiliate partnerDeepSeek R1 0528 Qwen3 8B on Novita AICall this model through Novita's API or try it in their playground, no GPU needed.Open model page →Affiliate partnerNovita SandboxCloud runtimes for AI agents: browser access, filesystem, computer use, long-running tasks.See Sandbox →

Don't want to host it yourself? Compare AI APIs and Sandboxes.

Or call it by API: compare LLM API prices.

See all GPU rental prices for more providers.

Where to rent or buy

Run it on a cloud GPU

  • Vast.ai Affiliate partner Consumer GPU marketplace (4090/3090-class rentals)
  • Runpod Affiliate partner Pods and serverless GPUs (referral link)
  • E2E Networks Works with India GPU cloud (official site)

Run it via API instead

  • Novita AI Affiliate partner Open-model APIs and Sandbox for AI agents (no GPU rental)

Buy this GPU (or a Mac)

  • B&H Photo Works with GPUs and computing gear (official site)

Which card? Read our best GPUs with 12 GB buying guide (the smallest size this model fits comfortably).

We may earn from links marked Affiliate partner. Disclosure.

Frequently asked questions

How much VRAM does DeepSeek-R1-0528-Qwen3-8B need at 4-bit?

About 5.0 GB for the Q4_K_M weights alone.[4] An 8K FP16 KV cache adds about 1.21 GB, for roughly 6.2 GB before engine overhead.[3]

Can I run it on a consumer GPU?

Yes, at Q4_K_M on an 8–12 GB card depending on context. Check the fit table above for your exact card.

Is BF16 practical?

BF16 is about 16.4 GB.[4] Prefer Q4_K_M or Q8_0 unless you have the memory.

How we calculated

  • Weights are the exact GGUF file sizes on Hugging Face (shards added together).[4]
  • KV cache is calculated from config.json at FP16.[3]
  • Device memory is from NVIDIA and Apple spec pages.
  • Not included: engine working buffers, the OS and display. Leave headroom.

Sources

All sources checked on .

  1. Hugging Face model page: deepseek-ai/DeepSeek-R1-0528-Qwen3-8B (parameter count, license, repo date). https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B
  2. Hugging Face model card / README for architecture and context. https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B
  3. config.json (deepseek-ai/DeepSeek-R1-0528-Qwen3-8B). https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/blob/main/config.json
  4. Hugging Face GGUF file sizes: bartowski/deepseek-ai_DeepSeek-R1-0528-Qwen3-8B-GGUF. https://huggingface.co/bartowski/deepseek-ai_DeepSeek-R1-0528-Qwen3-8B-GGUF/tree/main
  5. Ollama library tags (where listed). https://ollama.com/library/deepseek-r1
  6. Ollama pull / tag sizes where published for this family.
  7. NVIDIA GeForce RTX 4060 / 50-series 8 GB specs. https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5060-family/
  8. NVIDIA GeForce RTX 3060 specs (12 GB). https://www.nvidia.com/en-us/geforce/graphics-cards/30-series/rtx-3060-3060ti/
  9. NVIDIA GeForce RTX 4060 Ti / 16 GB class specs. https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/rtx-4060-4060ti/
  10. NVIDIA GeForce RTX 3090 specs (24 GB). https://www.nvidia.com/en-us/geforce/graphics-cards/30-series/rtx-3090-3090ti/
  11. NVIDIA GeForce RTX 4090 specs (24 GB). https://www.nvidia.com/en-us/geforce/graphics-cards/40-series/rtx-4090/
  12. NVIDIA GeForce RTX 5090 / RTX 6000 Ada specs. https://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/
  13. Apple MacBook Pro / Mac Studio unified memory specs. https://www.apple.com/macbook-pro/specs/
  14. Runpod, GPU pricing. https://www.runpod.io/pricing

Last checked: