GeekRadar / The comparison desk

COMPARE.

GPU

NVIDIA

NVIDIA RTX PRO 4000 Blackwell SFF

Graphics processor

Video memory: 24 GB^*
Memory bus: 192 bit^*
Rated memory bandwidth: 432 GB/s^*

No matched benchmark results are available for this entry yet.

Processing & platform

Architecture: Blackwell 2.0^*
Process node: 5 nm^*
GPU cores: 8,960^*
Graphics interface: PCIe 5.0 x8^*
Reported chip die area: 378 mm²^*
Transistors: 45,600 million^*
GPU L2 cache: 48 MB^*
Graphics processor: GB203^*
Texture mapping units: 280^*
Raster output units: 96^*
Streaming multiprocessors: 70^*
Tensor cores: 280^*
Ray tracing cores: 70^*
Chip foundry: TSMC^*
Card or module format: Dual-slot^*
Declared GPU TDP: 70 W^*
Suggested system PSU: 250 W^*
External power connectors: None^*
DirectX / feature level: 12 Ultimate (12_2)^*
OpenGL: 4.6^*
OpenCL: 3.0^*
Vulkan: 1.4^*
CUDA compute capability: 12.0^*
Shader model: 6.9^*

Memory & storage

Memory options: 24^*
Video memory: 24 GB^*
Video memory type: GDDR7^*
Memory bus: 192 bit^*
Rated memory bandwidth: 432 GB/s^*

Power specifications

Thermal design power: 70 W^*

System configuration

Detailed specifications

Specifications 1 · 12 fields

GPU architecture — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

Blackwell^*

Theoretical peak: FP32 non-tensor dense per GPU — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

24^* TFLOPS

Calculated theoretical peak: FP4 tensor dense per GPU — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

385^* TFLOPS

Scope and conditions

Calculation: 770 / 2^*

Theoretical peak: FP4 tensor with sparsity per GPU — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

770^* TFLOPS

Physical configuration — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

PCIe, half height, dual slot (2.7 in x 6.6 in L)^*

PCIe interface declaration — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

PCIe Gen 5 x8^*

Maximum board power — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

70^* W

Peak memory bandwidth per GPU — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

432^* GB/s

Memory interface width per GPU — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

192^* bit

Memory capacity per GPU — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

24^* GB

Memory technology — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

GDDR7^*

GPU model — NVIDIA RTX PRO 4000 Blackwell SFF Edition — 24 GB GDDR7

NVIDIA RTX PRO 4000 Blackwell SFF Edition^*

Specifications 2 · 30 fields

proc transistors: 45,600 million^*
proc die size: 378^* mm²
cores: 8,960^*
tensor cores: 280^*
rt cores: 70^*
sm count: 70^*
base clock: 577^*
boost clock: 1,432^*
memory clock: 1,125^*
ram: 24^*
name: NVIDIA RTX PRO 4000 Blackwell SFF^*
ram type: GDDR7^*
bus width: 192^*
bus interface: PCIe 5.0 x8^*
ram bandwidth: 432^* GB/s
fp16 performance: 25.66^*
fp32 performance: 25.66^*
fp64 performance: 0.401^*
tdp: 70^* W
suggested psu: 250^* W
slot width: Dual-slot^*
api cuda: 12^*
is retail board: No^*
gpi value: 13.27^*
vendor: nvd^*
manufacturer: NVIDIA^*
arch name: Blackwell 2.0^*
card release date: 2025-08-11^*
proc foundry: TSMC^*
proc process size: 5^* nm

Specifications 3 · 40 fields

CUDA compute capability

12.0^*

DirectX / feature level

12 Ultimate (12_2)^*

OpenCL

3.0^*

OpenGL

4.6^*

Shader model

6.9^*

Vulkan

1.4^*

Reported theoretical BF16 peak

48.1^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Graphics interface

PCIe 5.0 x8^*

Memory bus

192^* bit

Reported shader processors

8,960^*

Reported theoretical FP16 peak

24.05^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Reported theoretical FP32 peak

24.05^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Reported theoretical FP64 peak

0.3758^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Declared L1 cache configuration

128 KB (per SM)^*

Declared GPU L2 cache

48 MB^*

Reported display outputs

4x mini-DisplayPort 2.1b^*

Theoretical pixel fill rate

128.8 GPixel/s^*

External power connectors

None^*

Reported chip die area

378 mm²^*

Chip foundry

TSMC^*

Process node declaration

5 nm^*

Transistors

45.6 billion^*

Graphics processor

GB203^*

Reported video memory

24^* GB

Rated memory bandwidth

432.0 GB/s^*

Video memory type

GDDR7^*

Raster output units

96^*

Ray tracing cores

70^*

Reported shading units

8,960^*

Card or module format

Dual-slot^*

Streaming multiprocessors

70^*

Suggested system PSU

250 W^*

Declared GPU TDP

70 W^*

Tensor cores

280^*

Theoretical texture fill rate

375.8 GTexel/s^*

Reported theoretical TF32 peak

24.05^* TFLOPS

Scope and conditions

Arithmetic density: Not stated^*

Texture mapping units

280^*

GPU cores

8,960^*

Video memory

24^* GB

Video memory type

GDDR7^*

* Provisional data