deepseek-ai/DeepSeek-V3 VRAM requirements

685B params, Multi-head latent attention (MLA)

DeepSeek-V3 needs about 673 GB of GPU memory for FP8 weights. The smallest fitting setup at 32k context is 8× NVIDIA B200. It supports 64 concurrent conversations with SGLang defaults.

GPU requirements by context length

8k context
GPUGPUsConcurrent users
NVIDIA B200864
NVIDIA H200 SXM864
NVIDIA RTX PRO 6000 Blackwell Server Edition82
NVIDIA H2082
NVIDIA H100 SXM16 (verify server configuration)64
NVIDIA H100 NVL16 (verify server configuration)64
NVIDIA A100 80GB SXM16 (verify server configuration)64
NVIDIA L40Sno fit
32k context
GPUGPUsConcurrent users
NVIDIA B200864
NVIDIA H200 SXM828
NVIDIA H100 SXM16 (verify server configuration)46
NVIDIA H100 NVL16 (verify server configuration)64
NVIDIA A100 80GB SXM16 (verify server configuration)38
NVIDIA RTX PRO 6000 Blackwell Server Edition16 (verify server configuration)64
NVIDIA H2016 (verify server configuration)64
NVIDIA L40Sno fit
128k context
GPUGPUsConcurrent users
NVIDIA B200816
NVIDIA H200 SXM87
NVIDIA H100 SXM16 (verify server configuration)11
NVIDIA H100 NVL16 (verify server configuration)16
NVIDIA A100 80GB SXM16 (verify server configuration)9
NVIDIA RTX PRO 6000 Blackwell Server Edition16 (verify server configuration)17
NVIDIA H2016 (verify server configuration)17
NVIDIA L40Sno fit

FP8 weights, SGLang defaults, estimates.

Weights by precision

Weight memory
PrecisionWeightsSmallest fitting setup at 32k
FP8about 673 GB8x NVIDIA B200
INT4 (AWQ)about 370 GB4x NVIDIA B200

Model notes

Attention
Multi-head latent attention (MLA)
Architecture
Mixture-of-Experts
Layers
61
Hidden size
7,168
Vocabulary
129,280
Native precision
bfloat16

MLA keeps one compact shared KV copy per GPU, so adding GPUs does not shrink the cache, but the copy itself is small for a model this size.

Frequently asked questions

Will deepseek-ai/DeepSeek-V3 run on a single H100?
No, a single H100 SXM cannot hold deepseek-ai/DeepSeek-V3 at 32k context with FP8 weights. The model requires multiple GPUs.
What is the cheapest GPU setup for deepseek-ai/DeepSeek-V3?
The cheapest fitting setup at 32k context is 8x NVIDIA B200 with FP8 weights and SGLang defaults.
How much VRAM does deepseek-ai/DeepSeek-V3 need at 128k context?
At 128k context on the cheapest fitting setup, deepseek-ai/DeepSeek-V3 uses about 84 GB of VRAM per GPU for FP8 weights and about 4.6 GB per conversation for the KV cache. The total VRAM needed depends on the GPU count and parallelism configuration.
Can I run deepseek-ai/DeepSeek-V3 with INT4 quantization?
Yes, deepseek-ai/DeepSeek-V3 has INT4 weight figures. Its weights take about 370 GB in INT4 versus about 673 GB in FP8. INT4 can make the model fit on fewer GPUs, but check inference quality for your workload.

Want a different context length, GPU, or quantization? The calculator runs the same numbers live, preloaded with this model.

Open in calculator

Don't want to run this yourself? We deploy and operate it for you. Book a call →

Related models

deepseek-ai/DeepSeek-V4-Flash deepseek-ai/DeepSeek-V4-Pro google/gemma-4-31B-it