Tesla K40m and CMP 50HX - video card comparison

Tesla K40m

CMP 50HX

Architecture

Kepler (2012−2018)

Turing (2018−2021)

GPU code name

GK110B

TU102

Market segment

Workstation

Desktop

Release date

November 2013

June 2021

Interface

PCIe 3.0 x16

PCIe 3.0 x16

Supplementary power connectors

-

2x 8-pin

Boost clock speed

876 MHz

1545 MHz

Core clock speed

745 MHz

-

Floating-point performance

4,291 gflops

-

Manufacturing process technology

28 nm

12 nm

Number of transistors

7,080 million

18,600 million

Pipelines / CUDA cores

2880

3584

Power consumption (TDP)

245 Watt

250 Watt

Texture fill rate

210.2

296.6

Memory type

GDDR5

GDDR6

Maximum RAM amount

12 GB

10 GB

Memory bus width

384 Bit

320 Bit

Memory clock speed

6008 MHz

14 GB/s

Memory bandwidth

288.4 GB/s

560.0 GB/s

Display Connectors

No outputs

No outputs

DirectX

12 (11_1)

12 Ultimate (12_2)

CUDA

3.5

7.5

OpenCL

1.2

3.0

OpenGL

4.6

4.6

Shader Model

5.1

6.6

Vulkan

1.1.126

1.2