Skip to content

Deepseek Coder 1.3B Instruct VRAM requirements

How much VRAM Deepseek Coder 1.3B Instruct needs
4-bit weights: 778 MBWith 8K context: 2.8 GBWith 16K context: 4.3 GB8-bit, 8K context: 3.4 GB
Parameters
1.3B
Longest context
16K
Released
Oct 2023
License
Deepseek
Strongest at
Coding

Its licence (Deepseek) is not a standard open-source one: check the terms before commercial use.

How much VRAM Deepseek Coder 1.3B Instruct needs

Memory Deepseek Coder 1.3B Instruct needs at each precision and context length
PrecisionWeightsWith 8KWith 16K
16-bit2.5 GB4.6 GB6.1 GB
8-bit1.3 GB3.4 GB4.9 GB
6-bit1.0 GB3.1 GB4.6 GB
5-bit913 MB3.0 GB4.5 GB
4-bit778 MB2.8 GB4.3 GB
3-bit628 MB2.7 GB4.2 GB
2-bit482 MB2.5 GB4.0 GB
1.6-bit265 MB2.3 GB3.8 GB

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

GPUs and Macs that run Deepseek Coder 1.3B Instruct

The best way Deepseek Coder 1.3B Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit16K127 tok/s0.2 sPDF: Deepseek Coder 1.3B Instruct on RTX 4060 report
RTX 50608-bit16K198 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 5060 report
RTX 30708-bit16K198 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 3070 report
RTX 3080 10 GB8-bit16K305 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 3080 10 GB report
RTX 3060 12 GB8-bit16K163 tok/s0.2 sPDF: Deepseek Coder 1.3B Instruct on RTX 3060 12 GB report
Arc B5808-bit16K135 tok/s0.2 sPDF: Deepseek Coder 1.3B Instruct on Arc B580 report
RTX 40708-bit16K219 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 4070 report
RTX 4070 Super8-bit16K219 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 4070 Super report
RTX 50708-bit16K277 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 5070 report
RTX 4060 Ti 16 GB8-bit16K134 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit16K198 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit16K159 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on Arc A770 16 GB report
Radeon RX 9070 XT8-bit16K211 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit16K277 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 4070 Ti Super report
RTX 4080 Super8-bit16K298 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 4080 Super report
RTX 5070 Ti8-bit16K346 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 5070 Ti report
RTX 50808-bit16K364 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 5080 report
Radeon RX 7900 XTX8-bit16K281 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on Radeon RX 7900 XTX report
RTX 30908-bit16K358 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 3090 report
RTX 40908-bit16K378 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 4090 report
Radeon AI PRO R97008-bit16K211 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on Radeon AI PRO R9700 report
RTX 50908-bit16K554 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 5090 report
2× RTX 30908-bit16K358 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on 2× RTX 3090 report
2× RTX 40908-bit16K378 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit16K462 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit16K364 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX 6000 Ada report
L40S8-bit16K337 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on L40S report
2× RTX 50908-bit16K554 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on 2× RTX 5090 report
4× RTX 30908-bit16K358 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit16K554 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit16K64 tok/s1.8 sPDF: Deepseek Coder 1.3B Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit16K64 tok/s1.8 sPDF: Deepseek Coder 1.3B Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit16K64 tok/s1.8 sPDF: Deepseek Coder 1.3B Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit16K125 tok/s0.9 sPDF: Deepseek Coder 1.3B Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit16K125 tok/s0.9 sPDF: Deepseek Coder 1.3B Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit16K125 tok/s0.9 sPDF: Deepseek Coder 1.3B Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit16K191 tok/s0.5 sPDF: Deepseek Coder 1.3B Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit16K205 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit16K223 tok/s0.3 sPDF: Deepseek Coder 1.3B Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit16K223 tok/s0.3 sPDF: Deepseek Coder 1.3B Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit16K223 tok/s0.3 sPDF: Deepseek Coder 1.3B Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit16K273 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit16K99 tok/s0.3 sPDF: Deepseek Coder 1.3B Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit16K99 tok/s0.3 sPDF: Deepseek Coder 1.3B Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit16K139 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on RTX Spark report
DGX Spark8-bit16K128 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on DGX Spark report
Dual DGX Spark8-bit16K128 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on Dual DGX Spark report
DGX Station GB3008-bit16K1,006 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit16K554 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit16K554 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit16K554 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit16K598 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on A100 80 GB SXM report
H100 SXM8-bit16K769 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on H100 SXM report
4× H200 NVL8-bit16K889 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct on 4× H200 NVL report
8× H200 NVL8-bit16K889 tok/s0.1 sPDF: Deepseek Coder 1.3B Instruct 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 TheBloke/deepseek-coder-1.3b-instruct-GGUF
vLLM
vllm serve deepseek-ai/deepseek-coder-1.3b-instruct

Questions

How much VRAM does Deepseek Coder 1.3B Instruct need?

Deepseek Coder 1.3B Instruct needs 778 MB for its weights at 4-bit, 2.8 GB with an 8K context and 4.3 GB with 16K. At 8-bit, 3.4 GB.

What is the minimum VRAM for Deepseek Coder 1.3B Instruct?

About 2.7 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 Coder 1.3B Instruct?

1.3B parameters: 2.5 GB of weights at 16-bit, 778 MB at 4-bit.

Can Deepseek Coder 1.3B Instruct run on an RTX 4090?

Yes. At 8-bit it holds 16K of context and writes about 378 tokens a second.

More models