H100 CNX and L40S - video card comparison
H100 CNX
L40S
Architecture
Hopper (2022)
Ada Lovelace
GPU code name
GH100
AD102
Market segment
Desktop
Desktop
Release date
March 2023
October 2022
Interface
PCIe 5.0 x16
PCIe 4.0 x16
Supplementary power connectors
8-pin EPS
1x 16-pin
Boost clock speed
1845 MHz
2520 MHz
Manufacturing process technology
4 nm
5 nm
Number of transistors
80,000 million
76,300 million
Pipelines / CUDA cores
14592
18176
Power consumption (TDP)
350 Watt
300 Watt
Texture fill rate
841.3
1,431
Memory type
HBM2e
GDDR6
Maximum RAM amount
80 GB
48 GB
Memory bus width
5120 Bit
384 Bit
Memory clock speed
3.2 GB/s
18 GB/s
Memory bandwidth
2,039 GB/s
864.0 GB/s
Display Connectors
No outputs
1x HDMI 2.1, 3x DisplayPort 1.4a
HDMI
-
+
DirectX
N/A
12 Ultimate (12_2)
CUDA
9.0
8.9
OpenCL
3.0
3.0
OpenGL
N/A
4.6
Shader Model
N/A
6.7
Vulkan
N/A
1.3