Sharon AI logo – AI infrastructure and GPU compute
Create Account

NVIDIA L40s GPU

The NVIDIA L40S GPU Accelerator, built on the NVIDIA Ada Lovelace architecture, delivers cutting-edge graphics, compute, and AI performance for modern data center workloads. With 48GB of GDDR6 memory, third-generation RT cores, and fourth-generation Tensor cores, it is optimized for virtual workstations, AI training, rendering, and complex visual computing tasks, making it ideal for enterprises requiring scalable performance across a wide range of applications.
Contact Sales

Basic Product Information

NVIDIA L40 GPU

Product Name

NVIDIA L40s GPU

Architecture

NVIDIA Ada Lovelace

Memory

48GB GDDR6 with ECC

Compute Power

Up to 1,466 TOPS (FP8 Tensor Core, with sparsity)

Use Cases

Generative AI, LLM inference, LLM fine-tuning and small-model training, NVIDIA Omniverse Enterprise, rendering, 3D graphics, streaming, and video content

Price

From $0.99/hr

Contact Sales
Contact Sales

Key Advantages

48GB GDDR6 Memory

Ideal for memory-intensive tasks like 3D modeling and large-scale simulation.

Third-Generation RT Cores

Enhanced ray-tracing performance for lifelike designs and real-time animations.

Fourth-Generation Tensor Cores

Faster AI training with optimized TF32 and support for structural sparsity.

Data Center Ready

Designed for 24/7 enterprise data center operations and features secure boot with root of trust technology. It is also NEBS Level 3 ready.

vGPU Support

Distribute GPU resources efficiently with NVIDIA vGPU software for multiple users.

Specifications

Contact Sales

Performance Specifications

CUDA Cores

18,176

RT Cores

142 (Third-Generation)

Tensor Cores

568 (Fourth-Generation)

RT Core Performance

209 TFLOPS

FP32 Performance

91.6 TFLOPS

Tensor Core Performance:

TF32

183 TFLOPS | 366 TFLOPS (with sparsity)

BFLOAT16

362 TFLOPS | 733 TFLOPS (with sparsity)

FP16

362 TFLOPS | 733 TFLOPS (with sparsity)

FP8

733 TFLOPS | 1,466 TFLOPS (with sparsity)

INT8

733 TOPS | 1,466 TOPS (with sparsity)

INT4

733 TOPS | 1,466 TOPS (with sparsity)

Memory and Bandwidth

GPU Memory

48GB GDDR6 with ECC

Memory Bandwidth

864GB/s

Thermal and Power

Max Power Consumption

350W

Cooling

Passive cooling

Power Connector

16-pin

Board Specifications

Form Factor

Dual-slot (4.4” H x 10.5” L)

Interconnect Interface

PCIe Gen4 x16 (64GB/s bi-directional)

Display Outputs

4x DisplayPort 1.4a

NVENC/NVDEC

3x NVENC / 3x NVDEC (with AV1 encode & decode support)

Thermal Solution

Passive

vGPU Software Support

Yes (refer to Virtual GPU Licensing Guide for specific profiles)

Secure Boot with Root of Trust

Yes

NEBS Ready

Level 3

MIG Support

No

NVLink Support

No

Supported Technologies

Virtual GPU (vGPU)

Supported for multi-user environments

Secure Boot with Root of Trust

Yes

NEBS Ready

Level 3 compliant

NVLink and MIG Support

No

Server Compatibility

Virtual GPU (vGPU) Software Support: Supports vGPU 16.1 (R535 GA6) or later and NVIDIA Virtual Compute Server Edition. It also supports NVIDIA RTX™ Virtual Workstation vDWS, NVIDIA GRID® Virtual Applications (GRID vApps), and NVIDIA GRID Virtual PC (GRID vPC).

Contact Sales

Additional Features

01

CUDA Cores and Efficiency

Accelerated single-precision floating point (FP32) throughput and improved power efficiency significantly boost performance. The GPU also supports 16-bit math capabilities (BF16) for mixed-precision workloads.

02

Hardware-Accelerated Video

Features three dedicated video encoders (NVENC) and three dedicated video decoders (NVDEC), including support for AV1 encode and decode, which accelerates video processing workflows.

03

DLSS 3

This breakthrough frame-generation technology, which leverages deep learning and the Ada Lovelace architecture, boosts rendering performance and delivers higher frames per second.

04

NVIDIA Omniverse™

The L40S is part of the NVIDIA OVX™ server platform and is optimized to deliver high-performance graphics, ray tracing, and simulation for NVIDIA Omniverse™.

05

Transformer Engine

This technology intelligently scans transformer architecture neural networks and automatically recasts between FP8 and FP16 precisions to deliver faster AI performance and accelerate training and inference.

Want to learn more?

Let’s chat