Skip to content

Deepseek Coder 7B Instruct V1.5 VRAM requirements

How much VRAM Deepseek Coder 7B Instruct V1.5 needs
4-bit weights: 3.9 GBWith 4K context: 6.5 GB8-bit, 4K context: 9.5 GB
Parameters
6.9B
Longest context
4K
Released
Jan 2024
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 7B Instruct V1.5 needs

Memory Deepseek Coder 7B Instruct V1.5 needs at each precision and context length
PrecisionWeightsWith 4K
16-bit12.9 GB15.5 GB
8-bit6.8 GB9.5 GB
6-bit5.3 GB7.9 GB
5-bit4.6 GB7.2 GB
4-bit3.9 GB6.5 GB
3-bit3.1 GB5.8 GB
2-bit2.4 GB5.1 GB
1.6-bit1.3 GB4.0 GB

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

GPUs and Macs that run Deepseek Coder 7B Instruct V1.5

The best way Deepseek Coder 7B Instruct V1.5 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40605-bit4K40 tok/s0.9 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 4060 report
RTX 50605-bit4K64 tok/s0.7 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 5060 report
RTX 30705-bit4K64 tok/s0.7 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 3070 report
RTX 3080 10 GB8-bit4K72 tok/s0.5 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 3080 10 GB report
RTX 3060 12 GB8-bit4K35 tok/s1.1 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 3060 12 GB report
Arc B5808-bit4K31 tok/s0.7 sPDF: Deepseek Coder 7B Instruct V1.5 on Arc B580 report
RTX 40708-bit4K49 tok/s0.5 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 4070 report
RTX 4070 Super8-bit4K49 tok/s0.4 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 4070 Super report
RTX 50708-bit4K64 tok/s0.4 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 5070 report
RTX 4060 Ti 16 GB8-bit4K28 tok/s0.6 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit4K44 tok/s0.6 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit4K38 tok/s0.7 sPDF: Deepseek Coder 7B Instruct V1.5 on Arc A770 16 GB report
Radeon RX 9070 XT8-bit4K51 tok/s0.4 sPDF: Deepseek Coder 7B Instruct V1.5 on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit4K64 tok/s0.3 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 4070 Ti Super report
RTX 4080 Super8-bit4K70 tok/s0.3 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 4080 Super report
RTX 5070 Ti8-bit4K84 tok/s0.3 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 5070 Ti report
RTX 50808-bit4K90 tok/s0.2 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 5080 report
Radeon RX 7900 XTX8-bit4K74 tok/s0.6 sPDF: Deepseek Coder 7B Instruct V1.5 on Radeon RX 7900 XTX report
RTX 30908-bit4K88 tok/s0.4 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 3090 report
RTX 40908-bit4K94 tok/s0.2 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 4090 report
Radeon AI PRO R97008-bit4K51 tok/s0.4 sPDF: Deepseek Coder 7B Instruct V1.5 on Radeon AI PRO R9700 report
RTX 50908-bit4K156 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 5090 report
2× RTX 30908-bit4K88 tok/s0.2 sPDF: Deepseek Coder 7B Instruct V1.5 on 2× RTX 3090 report
2× RTX 40908-bit4K94 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit4K121 tok/s0.2 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit4K90 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX 6000 Ada report
L40S8-bit4K81 tok/s0.2 sPDF: Deepseek Coder 7B Instruct V1.5 on L40S report
2× RTX 50908-bit4K156 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on 2× RTX 5090 report
4× RTX 30908-bit4K88 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit4K156 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit4K14 tok/s8.7 sPDF: Deepseek Coder 7B Instruct V1.5 on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit4K14 tok/s8.7 sPDF: Deepseek Coder 7B Instruct V1.5 on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit4K14 tok/s8.7 sPDF: Deepseek Coder 7B Instruct V1.5 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit4K30 tok/s4.3 sPDF: Deepseek Coder 7B Instruct V1.5 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit4K30 tok/s4.3 sPDF: Deepseek Coder 7B Instruct V1.5 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit4K30 tok/s4.3 sPDF: Deepseek Coder 7B Instruct V1.5 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit4K52 tok/s2.2 sPDF: Deepseek Coder 7B Instruct V1.5 on M4 Max 128 GB report
M5 Max 128 GB8-bit4K58 tok/s0.6 sPDF: Deepseek Coder 7B Instruct V1.5 on M5 Max 128 GB report
M3 Ultra 96 GB8-bit4K65 tok/s1.2 sPDF: Deepseek Coder 7B Instruct V1.5 on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit4K65 tok/s1.2 sPDF: Deepseek Coder 7B Instruct V1.5 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit4K65 tok/s1.2 sPDF: Deepseek Coder 7B Instruct V1.5 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit4K89 tok/s0.4 sPDF: Deepseek Coder 7B Instruct V1.5 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit4K22 tok/s1.2 sPDF: Deepseek Coder 7B Instruct V1.5 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit4K22 tok/s1.2 sPDF: Deepseek Coder 7B Instruct V1.5 on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit4K30 tok/s0.4 sPDF: Deepseek Coder 7B Instruct V1.5 on RTX Spark report
DGX Spark8-bit4K27 tok/s0.4 sPDF: Deepseek Coder 7B Instruct V1.5 on DGX Spark report
Dual DGX Spark8-bit4K27 tok/s0.2 sPDF: Deepseek Coder 7B Instruct V1.5 on Dual DGX Spark report
DGX Station GB3008-bit4K437 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit4K156 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit4K156 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit4K156 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit4K174 tok/s0.3 sPDF: Deepseek Coder 7B Instruct V1.5 on A100 80 GB SXM report
H100 SXM8-bit4K260 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on H100 SXM report
4× H200 NVL8-bit4K339 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 on 4× H200 NVL report
8× H200 NVL8-bit4K339 tok/s0.1 sPDF: Deepseek Coder 7B Instruct V1.5 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 mradermacher/deepseek-coder-7b-instruct-v1.5-i1-GGUF
vLLM
vllm serve deepseek-ai/deepseek-coder-7b-instruct-v1.5

Questions

How much VRAM does Deepseek Coder 7B Instruct V1.5 need?

Deepseek Coder 7B Instruct V1.5 needs 3.9 GB for its weights at 4-bit, 6.5 GB with an 4K context and 6.5 GB with 4K. At 8-bit, 9.5 GB.

What is the minimum VRAM for Deepseek Coder 7B Instruct V1.5?

About 5.8 GB at 3-bit with an 4K 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 7B Instruct V1.5?

6.9B parameters: 12.9 GB of weights at 16-bit, 3.9 GB at 4-bit.

Can Deepseek Coder 7B Instruct V1.5 run on an RTX 4090?

Yes. At 8-bit it holds 4K of context and writes about 94 tokens a second.

More models