Skip to main content
  • This item is not returnable
Use the Auto-Notify button to be informed when this product becomes available.
Login to Make an Offer
Sold by Newegg

Shipped by Newegg

Server System configurator/ABS
Shop it at NeweggBusiness.

NVIDIA L40S Graphic Card - 48 GB GDDR6 - PCIe 4.0 x16 - 2x Slot - Ada - 900-2G133-0080-000

  • FP32: 91.6 teraFLOPS
  • TF32 Tensor Core: 366 teraFLOPS*
  • FP16: 733 teraFLOPS*
  • FP8: 1,466 teraFLOPS*
  • RT Core Performance: 212 teraFLOPS
  • Max Power Consumption: 350W
This item is covered under the Manufacturer Only Return Policy.
+
+
Specs

Any questions? Ask CelesteSM will help you find out quickly.

Learn more about the NVIDIA L40S

Best Seller Ranking #3 in AI Accelerator Cards
Model
Brand NVIDIA
Model L40S
Part Number 900-2G133-0080-000
Interface PCI Express 4.0 x16
Chipset
Chipset Manufacturer NVIDIA
GPU L40S
CUDA Cores 18176
Memory
Memory Size 48GB
Memory Clock 864 GB/s
Memory Type GDDR6 w/ECC
3D API
DirectX DirectX 12.07
Ports
DisplayPort 4
Form Factor & Dimensions
Dimensions (L x H) Height: 4.4"
Length:10.5"
Slot Width Dual
Additional Information
First Listed on Newegg August 06, 2024

Warranty & Returns

Warranty, Returns, And Additional Information

Warranty

  • Limited Warranty period (parts): 3 years
  • Limited Warranty period (labor): 3 years
  • Read full details

Return Policies

LOADING...
LOADING...

The NVIDIA L40S is a 48GB GDDR6 ECC, dual-slot PCIe 4.0 x16 accelerator for data-center AI inference, AI development, and graphics workloads. Its 18,176 CUDA cores and high Tensor Core throughput suit teams running compute-intensive workloads on compatible systems.

Recommended For

  • AI teams serving LLMs or image-generation models on a data-center server, when 48 GB of GPU memory fits the model and workload.
  • Teams running mixed AI and visual-computing workloads on shared infrastructure, such as inference alongside 3D rendering, graphics, or video processing.
  • Data-center and cloud operators providing virtual workstations or GPU-backed applications to multiple users, using supported NVIDIA vGPU software.
  • Model-development teams fine-tuning or testing workloads that fit within 48 GB per GPU and can use the L40S’s Tensor Core acceleration in a server deployment.

Spec Summary

AttributeValue
Best ForAI acceleration and graphics workloads, with an L40S GPU and 48 GB of ECC GDDR6 memory
GPU And ComputeNVIDIA L40S with 18,176 CUDA cores
Memory48 GB GDDR6 with ECC and 864 GB/s memory bandwidth
Interface And ExpansionPCI Express 4.0 x16 interface in a dual-slot design
Display Support4 DisplayPort outputs; supports DirectX 12.07
Physical Dimensions10.5 in. long and 4.4 in. high
Model IdentifiersNVIDIA L40S; part number 900-2G133-0080-000