NVIDIA L40S is a professional graphics card designed for data centers, with 48 GB of GDDR6 ECC memory in a passive (fanless) design. Based on the Ada Lovelace architecture (AD102), the L40S is intended for AI inference/training, 3D visualization, rendering, and video workflows.
- CUDA Cores: 18,176
- Tensor Cores: 568 (4th generation with FP8 support)
- RT Cores: 142 (3rd generation)
- L2 cache: 48 MB
- Memory interface: 384-bit with 864 GB/s bandwidth
- Boost clock: up to ~2520 MHz
- TDP: ~350 W
- Power connector: 12VHPWR (16-pin)
- Interface: PCIe x16 (PCIe Gen4 x16)
- Video outputs: 4× DisplayPort 1.4a
- Form factor: dual-slot, passive cooling
- Product Number
- 900-2G133-0080-000
Main Specifications
- Processor Manufacturer
- Nvidia
- Chipset Model
- NVIDIA L40S
- Interface
- PCI Express x16 4.0
- Maximum Resolution
- -
- Parallel Processing Technology
- Not Supported
- Type
- Professional
- Cuda Cores
- 18176
- Device Group
- Machine Learning, Rendering
- Colour
- Black
Special Features
- RGB Lighting
- No
- OverClock (OC)
- No
- Air-cooled
- No
- Low Profile
- No
- Passive Cooling
- Yes
- Water-cooled
- No
- External Graphics Card (eGPU)
- No
- LHR
- No
Memory
- Memory Capacity
- 48 GB
- Memory type
- GDDR6
- Memory Speed
- -
- Memory Speed
- 18 Gbps
- Memory bus
- 384 bit
Connectivity
- DisplayPort
- 4
- Dual Link DVI
- No
System Requirements
- Minimum Power Supply
- 350 W
- Power Connectors
- 16-pin
Top Specs
- Memory
- 48GB, GDDR6
- Connectivity
- DisplayPort
Important information
Specifications are collected from official manufacturer websites. Please verify the specifications before proceeding with your final purchase. If you notice any problem you can report it here.