GPU GUIDE · NVIDIA

Best AI models for the
NVIDIA DGX Station (Blackwell Ultra)

The NVIDIA DGX Station (Blackwell Ultra) has 784GB of unified memory. Below are the top 30 open-source AI models that fit, ranked by composite benchmark score. Each row shows the best quantization that fits your hardware.

VRAM
784GB
Brand
nvidia
Models that fit
30
Generation
DGX

Top 30 models for the NVIDIA DGX Station (Blackwell Ultra)

01

DeepSeek V4 Pro

685B
73.8
Best fit: Q8_0 · 728.8GBCan I run it? →
02

MiMo V2.5 Pro

1023B
70.4
Best fit: Q5_K_M · 727.5GBCan I run it? →
03

DeepSeek V4 Flash

37.0B
67.1
Best fit: fp16 · 75.0GBCan I run it? →
04

Qwen 3.6 Max

480B
66.7
Best fit: Q8_0 · 511.0GBCan I run it? →
05

Qwen 3.6 Plus

235B
65.9
Best fit: fp16 · 471.0GBCan I run it? →
06

MiniMax M2.7

229B
63.6
Best fit: fp16 · 458.4GBCan I run it? →
07

Qwen 3.6 27B

27.0B
61.8
Best fit: fp16 · 55.0GBCan I run it? →
08

MiMo V2 Omni

120B
60.7
Best fit: fp16 · 241.0GBCan I run it? →
09

NVIDIA Nemotron 3 Super

340B
60.0
Best fit: Q8_0 · 362.3GBCan I run it? →
10

Qwen3.5-27B

27.8B
56.3
Best fit: f32 · 112.2GBCan I run it? →
11

Qwen 3.5 397B A17B

397B
56.1
Best fit: Q8_0 · 422.8GBCan I run it? →
12

Qwen 3.5 122B A10B

122B
53.8
Best fit: fp16 · 245.0GBCan I run it? →
13

DeepSeek V3.2

685B
53.4
Best fit: Q8_0 · 728.8GBCan I run it? →
14

Qwen 3.6 35B A3B

35.0B
52.7
Best fit: fp16 · 71.0GBCan I run it? →
15

Qwen 3.5 Omni Plus

235B
51.1
Best fit: fp16 · 471.0GBCan I run it? →
16

Mistral Medium 3.5 128B

128B
49.9
Best fit: f32 · 511.8GBCan I run it? →
17

Gemma 4 31B

31.0B
48.9
Best fit: fp16 · 63.0GBCan I run it? →
18

Qwen 3.5 35B A3B

35.0B
48.8
Best fit: fp16 · 71.0GBCan I run it? →
19

Qwen 3 Coder

480B
46.7
Best fit: Q8_0 · 511.0GBCan I run it? →
20

Qwen 3 Next 80B A3B

80.0B
45.0
Best fit: fp16 · 161.0GBCan I run it? →
21

Gemma 4 26B A4B

26.5B
42.8
Best fit: f32 · 107.0GBCan I run it? →
22

NVIDIA Nemotron 3 Super 120B A12B BF16

124B
42.3
Best fit: fp16 · 248.2GBCan I run it? →
23

GPT-OSS 120B

120B
39.7
Best fit: fp16 · 241.8GBCan I run it? →
24

Devstral 2

24.0B
36.7
Best fit: fp16 · 49.0GBCan I run it? →
25

gemma 4 12B it

12.0B
36.4
Best fit: f32 · 49.0GBCan I run it? →
26

Qwen 3.5 9B

9.0B
35.7
Best fit: fp16 · 19.0GBCan I run it? →
27

Qwen3 VL 235B A22B Instruct

236B
34.3
Best fit: fp16 · 472.4GBCan I run it? →
28

Qwen 3.5 4B

4.0B
33.5
Best fit: fp16 · 9.0GBCan I run it? →
29

DeepSeek R1 0528

685B
33.5
Best fit: Q8_0 · 728.8GBCan I run it? →
30

Qwen3 235B A22B Thinking 2507

235B
32.7
Best fit: Q6_K · 194.7GBCan I run it? →
Advertisement

FAQ — running AI on the NVIDIA DGX Station (Blackwell Ultra)

How many AI models can the NVIDIA DGX Station (Blackwell Ultra) run?

With 784GB of unified memory, the NVIDIA DGX Station (Blackwell Ultra) can run 30+ open-source models from our database, including DeepSeek V4 Pro, MiMo V2.5 Pro, DeepSeek V4 Flash.

What's the largest LLM I can run on a NVIDIA DGX Station (Blackwell Ultra)?

The biggest model that fits is approximately 1023B. Larger models would need to be quantized further or won't fit at all.

Is 784GB of unified memory enough for local AI?

Yes — 784GB comfortably runs most popular open-source models including 30B-class LLMs at Q4_K_M.