L40S PCIe Gen4 x16

NVIDIA

NVIDIA · NVIDIA Ada Lovelace Architecture · Released August 2023 · 18,176 CUDA Cores · 48GB GDDR6 with ECC VRAM · 350W TDP

L40S PCIe Gen4 x16

OVERVIEW

The NVIDIA L40S is a high-performance datacenter GPU designed for AI, machine learning, and graphics-intensive workloads. It is part of the Ada Lovelace architecture, offering significant improvements in performance and efficiency over previous generations. Targeted at enterprise and cloud environments, the L40S excels in delivering accelerated computing power for demanding applications.

SPECIFICATIONS

RT Cores142
CUDA Cores18176
GPU Memory48GB GDDR6 with ECC
NEBS ReadyLevel 3
FP32 TFLOPS91.6
Form Factor4.4" (H) x 10.5" (L), dual slot
MIG SupportNo
Tensor Cores568
Display Ports4x DisplayPort 1.4a
NVENC | NVDEC3x | 3x (includes AV1 encode and decode)
NVLink SupportNo
FP8 Tensor Core733
Power Connector16-pin
GPU ArchitectureNVIDIA Ada Lovelace Architecture
Memory Bandwidth864GB/s
Max Power Consumption350W
Peak INT4 Tensor TOPS733
Peak INT8 Tensor TOPS733
Interconnect InterfacePCIe Gen4 x16: 64GB/s bidirectional
FP16 Tensor Core TFLOPS362.05
TF32 Tensor Core TFLOPS183
RT Core Performance TFLOPS209
BFLOAT16 Tensor Core TFLOPS362.05
Secure Boot With Root of TrustYes
Virtual GPU (vGPU) Software SupportYes

WHAT THIS GPU IS GOOD AT

The L40S excels in AI training and inference, high-performance computing, and rendering tasks. Its advanced architecture and enhanced tensor cores make it particularly effective for deep learning workloads, offering superior performance and efficiency. The GPU's capabilities in real-time ray tracing and graphics rendering also make it a strong choice for visual computing applications.

SERVER OPTIONS

The NVIDIA L40S is available in a variety of server platforms, including OEM servers like Dell PowerEdge, HPE ProLiant, and Supermicro. It is also offered in cloud instances such as AWS, Azure, and Google Cloud, providing flexible deployment options for enterprises seeking scalable AI and graphics solutions.

POWER, THERMALS & NOISE

The L40S has a TDP of around 300 watts, necessitating robust cooling solutions, typically air-cooled in datacenter environments. It maintains efficient thermal performance under load, though noise levels can vary depending on the server's cooling design. Liquid cooling options may be available for optimized thermal management.

COMPATIBILITY & SYSTEM FIT

The L40S uses a PCIe Gen4 x16 interface, ensuring compatibility with modern server motherboards. It supports NVLink for high-speed interconnects between multiple GPUs, enhancing scalability for large-scale deployments. Adequate power supply units are required to meet its power demands, typically around 750 watts or higher.

LIMITATIONS & KNOWN TRADE-OFFS

While the L40S offers impressive performance, it may be overkill for less demanding applications, leading to underutilization. Its high power consumption and cooling requirements can be a consideration for energy-conscious deployments. Availability may be limited initially due to high demand and production constraints.

NOTES

Unparalleled AI and graphics performance for the data center. Ideal for generative AI, large language model training, rendering, 3D graphics, and video content AI training.

"The L40S GPU is optimized for 24/7 enterprise data center operations and designed, built, tested, and supported by NVIDIA to ensure maximum performance, durability, and uptime. It meets the latest data center standards, is Network Equipment-Building System (NEBS) Level 3 ready, and features secure boot with root of trust technology."