Can I run DeepSeek R1 Distill Llama 8B locally?
DeepSeek R1 Distill Llama 8B by deepseek-ai has 8.0B parameters. At 4-bit with an 8K context it needs about 6.2 GB of GPU memory. It fits on 65 of 65 GPUs and Macs we track. See all models that fit in 8 GB of VRAM.
Memory needed by quant and context length
GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use real file sizes from unsloth/DeepSeek-R1-Distill-Llama-8B-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.
| Quant | Quality | 4K ctx | 8K ctx | 32K ctx | 128K ctx |
|---|---|---|---|---|---|
| 2-bit (Q2_K) • | rough | 4.1 GB | 4.6 GB | 7.6 GB | 19.6 GB |
| 3-bit (Q3_K_M) • | usable | 4.9 GB | 5.4 GB | 8.4 GB | 20.4 GB |
| 4-bit (Q4_K_M) • | good, most popular | 5.7 GB | 6.2 GB | 9.2 GB | 21.2 GB |
| 5-bit (Q5_K_M) • | very good | 6.5 GB | 7.0 GB | 10.0 GB | 22.0 GB |
| 6-bit (Q6_K) • | near lossless | 7.3 GB | 7.8 GB | 10.8 GB | 22.8 GB |
| 8-bit (Q8_0) • | effectively lossless | 9.1 GB | 9.6 GB | 12.6 GB | 24.6 GB |
| 16-bit (F16/BF16) • | original weights | 31.1 GB | 31.6 GB | 34.6 GB | 46.6 GB |
Which GPUs and Macs run DeepSeek R1 Distill Llama 8B?
Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).
| Hardware | Memory | Fits? | Best quant | Needs | Max tok/s (est.) | |
|---|---|---|---|---|---|---|
| RTX 3060 12GB | 12 GB | Fits | Q8_0 | 9.6 GB | 42 | details |
| RTX 4060 Ti 16GB | 16 GB | Fits | Q8_0 | 9.6 GB | 34 | details |
| RTX 3090 | 24 GB | Fits | Q8_0 | 9.6 GB | 110 | details |
| RTX 4090 | 24 GB | Fits | Q8_0 | 9.6 GB | 118 | details |
| RTX 5090 | 32 GB | Fits | Q8_0 | 9.6 GB | 210 | details |
| RTX 5070 Ti | 16 GB | Fits | Q8_0 | 9.6 GB | 105 | details |
| RX 7900 XTX | 24 GB | Fits | Q8_0 | 9.6 GB | 112 | details |
| Mac M4 Pro 48GB | 48 GB | Fits | Q8_0 | 9.6 GB | 32 | details |
| Mac M4 Max 128GB | 128 GB | Fits | F16 | 31.6 GB | 17 | details |
| Mac M3 Ultra 512GB | 512 GB | Fits | F16 | 31.6 GB | 25 | details |
| DGX Spark | 128 GB | Fits | F16 | 31.6 GB | 8.5 | details |
| RTX 4070 | 12 GB | Fits | Q8_0 | 9.6 GB | 59 | details |
| RTX 5060 Ti 16GB | 16 GB | Fits | Q8_0 | 9.6 GB | 52 | details |
| RTX 4060 | 8 GB | Fits | Q4_K_M | 6.2 GB | 55 | details |
| Arc B580 | 12 GB | Fits | Q8_0 | 9.6 GB | 53 | details |
| RTX 3060 Ti | 8 GB | Fits | Q4_K_M | 6.2 GB | 91 | details |
| RTX 3070 | 8 GB | Fits | Q4_K_M | 6.2 GB | 91 | details |
| RTX 3080 10GB | 10 GB | Fits | Q6_K | 7.8 GB | 115 | details |
| RTX 3080 12GB | 12 GB | Fits | Q8_0 | 9.6 GB | 107 | details |
| RTX 3090 Ti | 24 GB | Fits | Q8_0 | 9.6 GB | 118 | details |
| RTX 4060 Ti 8GB | 8 GB | Fits | Q4_K_M | 6.2 GB | 59 | details |
| RTX 4070 Super | 12 GB | Fits | Q8_0 | 9.6 GB | 59 | details |
| RTX 4070 Ti | 12 GB | Fits | Q8_0 | 9.6 GB | 59 | details |
| RTX 4070 Ti Super | 16 GB | Fits | Q8_0 | 9.6 GB | 79 | details |
| RTX 4080 | 16 GB | Fits | Q8_0 | 9.6 GB | 84 | details |
| RTX 4080 Super | 16 GB | Fits | Q8_0 | 9.6 GB | 86 | details |
| RTX 5060 | 8 GB | Fits | Q4_K_M | 6.2 GB | 91 | details |
| RTX 5070 | 12 GB | Fits | Q8_0 | 9.6 GB | 79 | details |
| RTX 5080 | 16 GB | Fits | Q8_0 | 9.6 GB | 112 | details |
| 2x RTX 3090 | 2×24 GB | Fits | F16 | 31.6 GB | 29 | details |
| 2x RTX 4090 | 2×24 GB | Fits | F16 | 31.6 GB | 31 | details |
| 2x RTX 5090 | 2×32 GB | Fits | F16 | 31.6 GB | 56 | details |
| RTX A6000 | 48 GB | Fits | F16 | 31.6 GB | 24 | details |
| RTX 6000 Ada | 48 GB | Fits | F16 | 31.6 GB | 30 | details |
| RTX PRO 6000 Blackwell | 96 GB | Fits | F16 | 31.6 GB | 56 | details |
| L40S | 48 GB | Fits | F16 | 31.6 GB | 27 | details |
| A100 40GB | 40 GB | Fits | F16 | 31.6 GB | 48 | details |
| A100 80GB | 80 GB | Fits | F16 | 31.6 GB | 63 | details |
| H100 PCIe | 80 GB | Fits | F16 | 31.6 GB | 62 | details |
| H100 SXM | 80 GB | Fits | F16 | 31.6 GB | 104 | details |
| H200 | 141 GB | Fits | F16 | 31.6 GB | 149 | details |
| RX 7600 | 8 GB | Fits | Q4_K_M | 6.2 GB | 59 | details |
| RX 7700 XT | 12 GB | Fits | Q8_0 | 9.6 GB | 51 | details |
| RX 7800 XT | 16 GB | Fits | Q8_0 | 9.6 GB | 73 | details |
| RX 7900 GRE | 16 GB | Fits | Q8_0 | 9.6 GB | 67 | details |
| RX 7900 XT | 20 GB | Fits | Q8_0 | 9.6 GB | 94 | details |
| RX 9070 | 16 GB | Fits | Q8_0 | 9.6 GB | 75 | details |
| RX 9070 XT | 16 GB | Fits | Q8_0 | 9.6 GB | 75 | details |
| Ryzen AI Max+ 395 128GB | 128 GB | Fits | F16 | 31.6 GB | 8.0 | details |
| Arc A770 16GB | 16 GB | Fits | Q8_0 | 9.6 GB | 66 | details |
| Mac M1 16GB | 16 GB | Fits | Q8_0 | 9.6 GB | 8.0 | details |
| Mac M2 24GB | 24 GB | Fits | Q8_0 | 9.6 GB | 12 | details |
| Mac M3 24GB | 24 GB | Fits | Q8_0 | 9.6 GB | 12 | details |
| Mac M4 16GB | 16 GB | Fits | Q8_0 | 9.6 GB | 14 | details |
| Mac M4 32GB | 32 GB | Fits | Q8_0 | 9.6 GB | 14 | details |
| Mac M1 Pro 32GB | 32 GB | Fits | Q8_0 | 9.6 GB | 23 | details |
| Mac M3 Pro 36GB | 36 GB | Fits | Q8_0 | 9.6 GB | 18 | details |
| Mac M4 Pro 24GB | 24 GB | Fits | Q8_0 | 9.6 GB | 32 | details |
| Mac M4 Pro 64GB | 64 GB | Fits | F16 | 31.6 GB | 8.5 | details |
| Mac M1 Max 64GB | 64 GB | Fits | F16 | 31.6 GB | 12 | details |
| Mac M2 Max 96GB | 96 GB | Fits | F16 | 31.6 GB | 12 | details |
| Mac M3 Max 128GB | 128 GB | Fits | F16 | 31.6 GB | 12 | details |
| Mac M4 Max 64GB | 64 GB | Fits | F16 | 31.6 GB | 17 | details |
| Mac M2 Ultra 192GB | 192 GB | Fits | F16 | 31.6 GB | 25 | details |
| Mac M3 Ultra 256GB | 256 GB | Fits | F16 | 31.6 GB | 25 | details |
Questions
How much VRAM does DeepSeek R1 Distill Llama 8B need?
About 6.2 GB at 4-bit (Q4_K_M) with an 8K context: 4.6 GB of weights (real GGUF file size), 1.0 GB of KV cache and 0.6 GB of runtime overhead. At 8-bit it needs about 9.6 GB.
What is the smallest GPU that runs DeepSeek R1 Distill Llama 8B?
At Q4_K_M and 8K context, the smallest single consumer GPU in our table that fits it is the RTX 4060 (8 GB).
Can I run DeepSeek R1 Distill Llama 8B on a Mac?
Yes. At Q4_K_M with 8K context it fits on a Mac M1 16GB or larger, using the default macOS GPU memory limit.
Similar models
Same family or a similar parameter count, with the VRAM each needs at Q4_K_M and 8K context.
DeepSeek R1 0528 Qwen3 8B
8.2B · 6.5 GB at 4-bitDeepSeek R1 Distill Qwen 7B
7.6B · 5.5 GB at 4-bitdeepseek coder 7b instruct v1.5
6.9B · 8.3 GB at 4-bitLlama 3.1 8B Instruct
8.0B · 6.2 GB at 4-bitLlama 3.1 8B
8.0B · 6.2 GB at 4-bitHermes 3 Llama 3.1 8B
8.0B · 6.2 GB at 4-bitModel facts
| Hugging Face repo | deepseek-ai/DeepSeek-R1-Distill-Llama-8B |
| Parameters | 8,030,261,248 (8.0B) |
| Layers / hidden size | 32 / 4096 |
| Attention | GQA; 32 heads, 8 KV heads, head dim 128 |
| KV cache per token (FP16) | 128.0 KiB |
| Max context | 131,072 tokens |
| License | mit |
| Released on HF | Jan 20, 2025 |
| Downloads (30d) / likes | 74,720 / 921 |
Sources: Hugging Face model API · config.json (deepseek-ai/DeepSeek-R1-Distill-Llama-8B) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.