Skip to content

DeepSeek V2 Chat VRAM requirements

How much VRAM DeepSeek V2 Chat needs
4-bit weights: 133 GBWith 8K context: 134 GBWith 32K context: 136 GB8-bit, 8K context: 235 GB
Parameters
236B
Active per token
21.4B
Longest context
160K
Released
Apr 2024
License
Deepseek
Strongest at
Not rated

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

How much VRAM DeepSeek V2 Chat needs

Memory DeepSeek V2 Chat needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit439 GB440 GB442 GB448 GB
8-bit233 GB235 GB236 GB243 GB
6-bit180 GB181 GB183 GB189 GB
5-bit156 GB157 GB159 GB165 GB
4-bit133 GB134 GB136 GB142 GB
3-bit107 GB109 GB110 GB117 GB
2-bit82.3 GB83.7 GB85.2 GB91.6 GB
1.6-bit45.3 GB46.6 GB48.2 GB54.5 GB

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

GPUs and Macs that run DeepSeek V2 Chat

The best way DeepSeek V2 Chat runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: DeepSeek V2 Chat on RTX 4060 report
RTX 5060Too bigPDF: DeepSeek V2 Chat on RTX 5060 report
RTX 3070Too bigPDF: DeepSeek V2 Chat on RTX 3070 report
RTX 3080 10 GBToo bigPDF: DeepSeek V2 Chat on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: DeepSeek V2 Chat on RTX 3060 12 GB report
Arc B580Too bigPDF: DeepSeek V2 Chat on Arc B580 report
RTX 4070Too bigPDF: DeepSeek V2 Chat on RTX 4070 report
RTX 4070 SuperToo bigPDF: DeepSeek V2 Chat on RTX 4070 Super report
RTX 5070Too bigPDF: DeepSeek V2 Chat on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: DeepSeek V2 Chat on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: DeepSeek V2 Chat on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: DeepSeek V2 Chat on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: DeepSeek V2 Chat on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: DeepSeek V2 Chat on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: DeepSeek V2 Chat on RTX 4080 Super report
RTX 5070 TiToo bigPDF: DeepSeek V2 Chat on RTX 5070 Ti report
RTX 5080Too bigPDF: DeepSeek V2 Chat on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: DeepSeek V2 Chat on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: DeepSeek V2 Chat on RTX 3090 report
RTX 4090Too bigPDF: DeepSeek V2 Chat on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: DeepSeek V2 Chat on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: DeepSeek V2 Chat on RTX 5090 report
2× RTX 3090Too bigPDF: DeepSeek V2 Chat on 2× RTX 3090 report
2× RTX 4090Too bigPDF: DeepSeek V2 Chat on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: DeepSeek V2 Chat on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: DeepSeek V2 Chat on RTX 6000 Ada report
L40SToo bigPDF: DeepSeek V2 Chat on L40S report
2× RTX 5090Too bigPDF: DeepSeek V2 Chat on 2× RTX 5090 report
4× RTX 30902-bit88K63 tok/s1.3 sPDF: DeepSeek V2 Chat on 4× RTX 3090 report
RTX PRO 6000 Blackwell2-bit120K100 tok/s1.1 sPDF: DeepSeek V2 Chat on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: DeepSeek V2 Chat on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: DeepSeek V2 Chat on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: DeepSeek V2 Chat on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: DeepSeek V2 Chat on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: DeepSeek V2 Chat on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: DeepSeek V2 Chat on Mac mini M4 Pro 64 GB report
M4 Max 128 GB2-bit120K36 tok/s24 sPDF: DeepSeek V2 Chat on M4 Max 128 GB report
M5 Max 128 GB2-bit120K39 tok/s6.9 sPDF: DeepSeek V2 Chat on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: DeepSeek V2 Chat on M3 Ultra 96 GB report
M3 Ultra 256 GB6-bit24K24 tok/s13 sPDF: DeepSeek V2 Chat on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit160K20 tok/s13 sPDF: DeepSeek V2 Chat on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit160K27 tok/s4.2 sPDF: DeepSeek V2 Chat on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: DeepSeek V2 Chat on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB2-bit120K17 tok/s14 sPDF: DeepSeek V2 Chat on Ryzen AI Max+ 395 128 GB report
RTX Spark3-bit96K19 tok/s4.8 sPDF: DeepSeek V2 Chat on RTX Spark report
DGX Spark3-bit96K17 tok/s4.8 sPDF: DeepSeek V2 Chat on DGX Spark report
Dual DGX Spark6-bit160K10 tok/s2.8 sPDF: DeepSeek V2 Chat on Dual DGX Spark report
DGX Station GB3008-bit80K122 tok/s0.4 sPDF: DeepSeek V2 Chat on DGX Station GB300 report
2× RTX PRO 6000 Blackwell5-bit160K64 tok/s0.7 sPDF: DeepSeek V2 Chat on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit160K46 tok/s0.4 sPDF: DeepSeek V2 Chat on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit160K46 tok/s0.2 sPDF: DeepSeek V2 Chat on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: DeepSeek V2 Chat on A100 80 GB SXM report
H100 SXMToo bigPDF: DeepSeek V2 Chat on H100 SXM report
4× H200 NVL8-bit160K97 tok/s0.4 sPDF: DeepSeek V2 Chat on 4× H200 NVL report
8× H200 NVL8-bit160K97 tok/s0.2 sPDF: DeepSeek V2 Chat 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-V2-Chat-i1-GGUF
vLLM
vllm serve deepseek-ai/DeepSeek-V2-Chat

Questions

How much VRAM does DeepSeek V2 Chat need?

DeepSeek V2 Chat needs 133 GB for its weights at 4-bit, 134 GB with an 8K context and 136 GB with 32K. At 8-bit, 235 GB.

What is the minimum VRAM for DeepSeek V2 Chat?

About 83.7 GB at 2-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 4× RTX 3090.

How big is DeepSeek V2 Chat?

236B parameters (21.4B active per token): 439 GB of weights at 16-bit, 133 GB at 4-bit.

Can DeepSeek V2 Chat run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the 4× RTX 3090.

More models