Skip to content

DeepSeek R1 Distill Qwen 1.5B VRAM requirements

How much VRAM DeepSeek R1 Distill Qwen 1.5B needs
4-bit weights: 1.0 GBWith 8K context: 2.0 GBWith 32K context: 2.7 GB8-bit, 8K context: 2.8 GB
Parameters
1.8B
Longest context
128K
Released
Jan 2025
License
MIT
Strongest at
Not rated

How much VRAM DeepSeek R1 Distill Qwen 1.5B needs

Memory DeepSeek R1 Distill Qwen 1.5B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit3.3 GB4.3 GB5.0 GB7.6 GB
8-bit1.8 GB2.8 GB3.4 GB6.0 GB
6-bit1.4 GB2.4 GB3.0 GB5.6 GB
5-bit1.2 GB2.2 GB2.8 GB5.5 GB
4-bit1.0 GB2.0 GB2.7 GB5.3 GB
3-bit828 MB1.8 GB2.5 GB5.1 GB
2-bit636 MB1.6 GB2.3 GB4.9 GB
1.6-bit350 MB1.3 GB2.0 GB4.6 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run DeepSeek R1 Distill Qwen 1.5B

The best way DeepSeek R1 Distill Qwen 1.5B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit128K106 tok/s0.2 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 4060 report
RTX 50608-bit128K165 tok/s0.2 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 5060 report
RTX 30708-bit128K165 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 3070 report
RTX 3080 10 GB8-bit128K255 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit128K136 tok/s0.2 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 3060 12 GB report
Arc B5808-bit128K113 tok/s0.2 sPDF: DeepSeek R1 Distill Qwen 1.5B on Arc B580 report
RTX 40708-bit128K182 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 4070 report
RTX 4070 Super8-bit128K182 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 4070 Super report
RTX 50708-bit128K231 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit128K111 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit128K165 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit128K133 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit128K176 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit128K231 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 4070 Ti Super report
RTX 4080 Super8-bit128K249 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 4080 Super report
RTX 5070 Ti8-bit128K289 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 5070 Ti report
RTX 50808-bit128K305 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 5080 report
Radeon RX 7900 XTX8-bit128K235 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on Radeon RX 7900 XTX report
RTX 30908-bit128K299 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 3090 report
RTX 40908-bit128K316 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 4090 report
Radeon AI PRO R97008-bit128K176 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on Radeon AI PRO R9700 report
RTX 50908-bit128K466 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 5090 report
2× RTX 30908-bit128K299 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on 2× RTX 3090 report
2× RTX 40908-bit128K316 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit128K387 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit128K305 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX 6000 Ada report
L40S8-bit128K282 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on L40S report
2× RTX 50908-bit128K466 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on 2× RTX 5090 report
4× RTX 30908-bit128K299 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit128K466 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit128K53 tok/s1.9 sPDF: DeepSeek R1 Distill Qwen 1.5B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit128K53 tok/s1.9 sPDF: DeepSeek R1 Distill Qwen 1.5B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit128K53 tok/s1.9 sPDF: DeepSeek R1 Distill Qwen 1.5B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit128K105 tok/s1.0 sPDF: DeepSeek R1 Distill Qwen 1.5B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit128K105 tok/s1.0 sPDF: DeepSeek R1 Distill Qwen 1.5B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit128K105 tok/s1.0 sPDF: DeepSeek R1 Distill Qwen 1.5B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit128K160 tok/s0.5 sPDF: DeepSeek R1 Distill Qwen 1.5B on M4 Max 128 GB report
M5 Max 128 GB8-bit128K172 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit128K188 tok/s0.3 sPDF: DeepSeek R1 Distill Qwen 1.5B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit128K188 tok/s0.3 sPDF: DeepSeek R1 Distill Qwen 1.5B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit128K188 tok/s0.3 sPDF: DeepSeek R1 Distill Qwen 1.5B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit128K230 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit128K83 tok/s0.3 sPDF: DeepSeek R1 Distill Qwen 1.5B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit128K83 tok/s0.3 sPDF: DeepSeek R1 Distill Qwen 1.5B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit128K116 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on RTX Spark report
DGX Spark8-bit128K106 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on DGX Spark report
Dual DGX Spark8-bit128K106 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on Dual DGX Spark report
DGX Station GB3008-bit128K855 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit128K466 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit128K466 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit128K466 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit128K503 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on A100 80 GB SXM report
H100 SXM8-bit128K650 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on H100 SXM report
4× H200 NVL8-bit128K753 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on 4× H200 NVL report
8× H200 NVL8-bit128K753 tok/s0.1 sPDF: DeepSeek R1 Distill Qwen 1.5B on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf unsloth/DeepSeek-R1-Distill-Qwen-1.5B-GGUF
vLLM
vllm serve deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
MLX (Mac)
mlx_lm.chat --model mlx-community/DeepSeek-R1-Distill-Qwen-1.5B-4bit

Questions

How much VRAM does DeepSeek R1 Distill Qwen 1.5B need?

DeepSeek R1 Distill Qwen 1.5B needs 1.0 GB for its weights at 4-bit, 2.0 GB with an 8K context and 2.7 GB with 32K. At 8-bit, 2.8 GB.

What is the minimum VRAM for DeepSeek R1 Distill Qwen 1.5B?

About 1.8 GB at 3-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the RTX 4060.

How big is DeepSeek R1 Distill Qwen 1.5B?

1.8B parameters: 3.3 GB of weights at 16-bit, 1.0 GB at 4-bit.

Can DeepSeek R1 Distill Qwen 1.5B run on an RTX 4090?

Yes. At 8-bit it holds 128K of context and writes about 316 tokens a second.

More models