Skip to content

DeepSeek V4 Pro VRAM requirements

How much VRAM DeepSeek V4 Pro needs
FP4 + FP8 weights: 805 GBWith 8K context: 806 GBWith 32K context: 807 GB
Parameters
1.6T
Active per token
49B
Longest context
1M
Released
Apr 2026
License
MIT
Strongest at
Chat, Coding, Agents, Docs

How much VRAM DeepSeek V4 Pro needs

Memory DeepSeek V4 Pro needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
As released (FP4 + FP8)805 GB806 GB807 GB807 GB
3-bit728 GB729 GB729 GB730 GB
2-bit559 GB560 GB560 GB561 GB
1.6-bit307 GB308 GB309 GB309 GB

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

GPUs and Macs that run DeepSeek V4 Pro

The best way DeepSeek V4 Pro runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: DeepSeek V4 Pro on RTX 4060 report
RTX 5060Too bigPDF: DeepSeek V4 Pro on RTX 5060 report
RTX 3070Too bigPDF: DeepSeek V4 Pro on RTX 3070 report
RTX 3080 10 GBToo bigPDF: DeepSeek V4 Pro on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: DeepSeek V4 Pro on RTX 3060 12 GB report
Arc B580Too bigPDF: DeepSeek V4 Pro on Arc B580 report
RTX 4070Too bigPDF: DeepSeek V4 Pro on RTX 4070 report
RTX 4070 SuperToo bigPDF: DeepSeek V4 Pro on RTX 4070 Super report
RTX 5070Too bigPDF: DeepSeek V4 Pro on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: DeepSeek V4 Pro on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: DeepSeek V4 Pro on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: DeepSeek V4 Pro on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: DeepSeek V4 Pro on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: DeepSeek V4 Pro on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: DeepSeek V4 Pro on RTX 4080 Super report
RTX 5070 TiToo bigPDF: DeepSeek V4 Pro on RTX 5070 Ti report
RTX 5080Too bigPDF: DeepSeek V4 Pro on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: DeepSeek V4 Pro on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: DeepSeek V4 Pro on RTX 3090 report
RTX 4090Too bigPDF: DeepSeek V4 Pro on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: DeepSeek V4 Pro on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: DeepSeek V4 Pro on RTX 5090 report
2× RTX 3090Too bigPDF: DeepSeek V4 Pro on 2× RTX 3090 report
2× RTX 4090Too bigPDF: DeepSeek V4 Pro on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: DeepSeek V4 Pro on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: DeepSeek V4 Pro on RTX 6000 Ada report
L40SToo bigPDF: DeepSeek V4 Pro on L40S report
2× RTX 5090Too bigPDF: DeepSeek V4 Pro on 2× RTX 5090 report
4× RTX 3090Too bigPDF: DeepSeek V4 Pro on 4× RTX 3090 report
RTX PRO 6000 BlackwellToo bigPDF: DeepSeek V4 Pro on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: DeepSeek V4 Pro on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: DeepSeek V4 Pro on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: DeepSeek V4 Pro on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: DeepSeek V4 Pro on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: DeepSeek V4 Pro on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: DeepSeek V4 Pro on Mac mini M4 Pro 64 GB report
M4 Max 128 GBToo bigPDF: DeepSeek V4 Pro on M4 Max 128 GB report
M5 Max 128 GBToo bigPDF: DeepSeek V4 Pro on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: DeepSeek V4 Pro on M3 Ultra 96 GB report
M3 Ultra 256 GBToo bigPDF: DeepSeek V4 Pro on M3 Ultra 256 GB report
M3 Ultra 512 GB1.6-bit1M36 tok/s30 sPDF: DeepSeek V4 Pro on M3 Ultra 512 GB report
M5 Ultra 512 GB1.6-bit1M47 tok/s9.5 sPDF: DeepSeek V4 Pro on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: DeepSeek V4 Pro on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBToo bigPDF: DeepSeek V4 Pro on Ryzen AI Max+ 395 128 GB report
RTX SparkToo bigPDF: DeepSeek V4 Pro on RTX Spark report
DGX SparkToo bigPDF: DeepSeek V4 Pro on DGX Spark report
Dual DGX SparkToo bigPDF: DeepSeek V4 Pro on Dual DGX Spark report
DGX Station GB3002-bit, spills1M29 tok/s4.4 sPDF: DeepSeek V4 Pro on DGX Station GB300 report
2× RTX PRO 6000 BlackwellToo bigPDF: DeepSeek V4 Pro on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell1.6-bit1M85 tok/s0.8 sPDF: DeepSeek V4 Pro on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell2-bit1M55 tok/s0.4 sPDF: DeepSeek V4 Pro on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: DeepSeek V4 Pro on A100 80 GB SXM report
H100 SXMToo bigPDF: DeepSeek V4 Pro on H100 SXM report
4× H200 NVL1.6-bit1M150 tok/s0.8 sPDF: DeepSeek V4 Pro on 4× H200 NVL report
8× H200 NVLFP4 + FP81M64 tok/s0.4 sPDF: DeepSeek V4 Pro 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-V4-Pro-0813-GGUF
vLLM
vllm serve deepseek-ai/DeepSeek-V4-Pro
MLX (Mac)
mlx_lm.chat --model mlx-community/DeepSeek-V4-Pro-4bit

Questions

How much VRAM does DeepSeek V4 Pro need?

DeepSeek V4 Pro needs 805 GB for its weights as released (FP4 + FP8), 806 GB with an 8K context and 807 GB with 32K.

What is the minimum VRAM for DeepSeek V4 Pro?

About 308 GB at 1.6-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 M3 Ultra 512 GB.

How big is DeepSeek V4 Pro?

1.6T parameters (49B active per token): 805 GB of weights as released (FP4 + FP8).

Can DeepSeek V4 Pro run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the M3 Ultra 512 GB.

More models