L40 CNX
General
ManufacturerNVIDIA
SeriesServer Ada(Lxx)
GPU NameAD102
ArchitectureAda Lovelace
GenerationServer Ada(Lxx)
Release DateOct 2022
Bus InterfacePCIe 4.0 x16
Process & Die
FoundryTSMC
Process Size5 nm
Transistors76,300 M
Transistor Density125,300 K/mm²
Die Size609 mm²
Clock Speeds
Base Clock1,005 MHz
Boost Clock2,475 MHz
Memory
VRAM24 GB
Memory TypeGDDR6
Memory Clock2,250 MHz
Memory Bus384-bit
Bandwidth864 GB/s
Processing Units
Shading Units18,176
TMUs568
ROPs192
SMs142
Tensor Cores568
RT Cores142
Cache
L1 Cache128 KB
L2 Cache48 MB
Power & Thermals
TDP300 W
Suggested PSU700 W
Power Connectors1x 16-pin
Dimensions
Board Length267 mm
Board Width111 mm
Outputs
Display Connectors1x HDMI 2.1, 3x DisplayPort 1.4a
API Support
DirectX12.2
OpenGL4.6
Vulkan1.3
OpenCL3.0
CUDA8.9
Shader Model6.8
Theoretical Performance
Pixel Rate475.2 GP/s
Texture Rate1,405.8 GT/s
FP16 (Half)89,970 GFLOPS
FP32 (Single)89,970 GFLOPS
FP64 (Double)1,405.8 GFLOPS
External
TechPowerUpView →
Other Server Ada(Lxx) Models