LogoLogo
Continuum WebsiteContinuum ApplicationsContinuum KnowledgeAxolotl Platform
Continuum Knowledge
Continuum Knowledge
  • Continuum
  • Data
    • Datasets
      • Pre Training Data
      • Types of Fine Tuning
      • Self Instruct Paper
      • Self-Alignment with Instruction Backtranslation
      • Systematic Evaluation of Instruction-Tuned Large Language Models on Open Datasets
      • Instruction Tuning
      • Instruction Fine Tuning - Alpagasus
      • Less is More For Alignment
      • Enhanced Supervised Fine Tuning
      • Visualising Data using t-SNE
      • UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
      • Training and Evaluation Datasets
      • What is perplexity?
  • MODELS
    • Foundation Models
      • The leaderboard
      • Foundation Models
      • LLama 2 - Analysis
      • Analysis of Llama 3
      • Llama 3.1 series
      • Google Gemini 1.5
      • Platypus: Quick, Cheap, and Powerful Refinement of LLMs
      • Mixtral of Experts
      • Mixture-of-Agents (MoA)
      • Phi 1.5
        • Refining the Art of AI Training: A Deep Dive into Phi 1.5's Innovative Approach
      • Phi 2.0
      • Phi-3 Technical Report
  • Training
    • The Fine Tuning Process
      • Why fine tune?
        • Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations?
        • Explanations in Fine Tuning
      • Tokenization
        • Tokenization Is More Than Compression
        • Tokenization - SentencePiece
        • Tokenization explore
        • Tokenizer Choice For LLM Training: Negligible or Crucial?
        • Getting the most out of your tokenizer for pre-training and domain adaptation
        • TokenMonster
      • Parameter Efficient Fine Tuning
        • P-Tuning
          • The Power of Scale for Parameter-Efficient Prompt Tuning
        • Prefix-Tuning: Optimizing Continuous Prompts for Generation
        • Harnessing the Power of PEFT: A Smarter Approach to Fine-tuning Pre-trained Models
        • What is Low-Rank Adaptation (LoRA) - explained by the inventor
        • Low Rank Adaptation (Lora)
        • Practical Tips for Fine-tuning LMs Using LoRA (Low-Rank Adaptation)
        • QLORA: Efficient Finetuning of Quantized LLMs
        • Bits and Bytes
        • The Magic behind Qlora
        • Practical Guide to LoRA: Tips and Tricks for Effective Model Adaptation
        • The quantization constant
        • QLORA: Efficient Finetuning of Quantized Language Models
        • QLORA and Fine-Tuning of Quantized Language Models (LMs)
        • ReLoRA: High-Rank Training Through Low-Rank Updates
        • SLoRA: Federated Parameter Efficient Fine-Tuning of Language Models
        • GaLora: Memory-Efficient LLM Training by Gradient Low-Rank Projection
      • Hyperparameters
        • Batch Size
        • Padding Tokens
        • Mixed precision training
        • FP8 Formats for Deep Learning
        • Floating Point Numbers
        • Batch Size and Model loss
        • Batch Normalisation
        • Rethinking Learning Rate Tuning in the Era of Language Models
        • Sample Packing
        • Gradient accumulation
        • A process for choosing the learning rate
        • Learning Rate Scheduler
        • Checkpoints
        • A Survey on Efficient Training of Transformers
        • Sequence Length Warmup
        • Understanding Training vs. Evaluation Data Splits
        • Cross-entropy loss
        • Weight Decay
        • Optimiser
        • Caching
      • Training Processes
        • Extending the context window
        • PyTorch Fully Sharded Data Parallel (FSDP)
        • Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation
        • YaRN: Efficient Context Window Extension of Large Language Models
        • Sliding Window Attention
        • LongRoPE
        • Reinforcement Learning
        • An introduction to reinforcement learning
        • Reinforcement Learning from Human Feedback (RLHF)
        • Direct Preference Optimization: Your Language Model is Secretly a Reward Model
  • INFERENCE
    • Why is inference important?
      • Grouped Query Attention
      • Key Value Cache
      • Flash Attention
      • Flash Attention 2
      • StreamingLLM
      • Paged Attention and vLLM
      • TensorRT-LLM
      • Torchscript
      • NVIDIA L40S GPU
      • Triton Inference Server - Introduction
      • Triton Inference Server
      • FiDO: Fusion-in-Decoder optimised for stronger performance and faster inference
      • Is PUE a useful measure of data centre performance?
      • SLORA
  • KNOWLEDGE
    • Vector Databases
      • A Comprehensive Survey on Vector Databases
      • Vector database management systems: Fundamental concepts, use-cases, and current challenges
      • Using the Output Embedding to Improve Language Models
      • Decoding Sentence-BERT
      • ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT
      • SimCSE: Simple Contrastive Learning of Sentence Embeddings
      • Questions Are All You Need to Train a Dense Passage Retriever
      • Improving Text Embeddings with Large Language Models
      • Massive Text Embedding Benchmark
      • RocketQAv2: A Joint Training Method for Dense Passage Retrieval and Passage Re-ranking
      • LLM2Vec: Large Language Models Are Secretly Powerful Text Encoders
      • Embedding and Fine-Tuning in Neural Language Models
      • Embedding Model Construction
      • Demystifying Embedding Spaces using Large Language Models
      • Fine-Tuning Llama for Multi-Stage Text Retrieval
      • Large Language Model Based Text Augmentation Enhanced Personality Detection Model
      • One Embedder, Any Task: Instruction-Finetuned Text Embeddings
      • Vector Databases are not the only solution
      • Knowledge Graphs
        • Harnessing Knowledge Graphs to Elevate AI: A Technical Exploration
        • Unifying Large Language Models and Knowledge Graphs: A Roadmap
      • Approximate Nearest Neighbor (ANN)
      • High Dimensional Data
      • Principal Component Analysis (PCA)
      • Vector Similarity Search - HNSW
      • FAISS (Facebook AI Similarity Search)
      • Unsupervised Dense Retrievers
    • Retrieval Augmented Generation
      • Retrieval-Augmented Generation for Large Language Models: A Survey
      • Fine-Tuning or Retrieval?
      • Revolutionising Information Retrieval: The Power of RAG in Language Models
      • A Survey on Retrieval-Augmented Text Generation
      • REALM: Retrieval-Augmented Language Model Pre-Training
      • Retrieve Anything To Augment Large Language Models
      • Generate Rather Than Retrieve: Large Language Models Are Strong Context Generators
      • Active Retrieval Augmented Generation
      • DSPy: LM Assertions: Enhancing Language Model Pipelines with Computational Constraints
      • DSPy: Compiling Declarative Language Model Calls
      • DSPy: In-Context Learning for Extreme Multi-Label Classification
      • Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs
      • HYDE: Revolutionising Search with Hypothetical Document Embeddings
      • Enhancing Recommender Systems with Large Language Model Reasoning Graphs
      • Retrieval Augmented Generation (RAG) versus fine tuning
      • RAFT: Adapting Language Model to Domain Specific RAG
      • Summarisation Methods and RAG
      • Lessons Learned on LLM RAG Solutions
      • Stanford: Retrieval Augmented Language Models
      • Overview of RAG Approaches with Vector Databases
      • Mastering Chunking in Retrieval-Augmented Generation (RAG) Systems
    • Semantic Routing
    • Resource Description Framework (RDF)
  • AGENTS
    • What is agency?
      • Rephrase and Respond: Let Large Language Models Ask Better Questions for Themselves
      • Types of Agents
      • The risk of AI agency
      • Understanding Personality in Large Language Models: A New Frontier in AI Psychology
      • AI Agents - Reasoning, Planning, and Tool Calling
      • Personality and Brand
      • Agent Interaction via APIs
      • Bridging Minds and Machines: The Legacy of Newell, Shaw, and Simon
      • A Survey on Language Model based Autonomous Agents
      • Large Language Models as Agents
      • AI Reasoning: A Deep Dive into Chain-of-Thought Prompting
      • Enhancing AI Reasoning with Self-Taught Reasoner (STaR)
      • Exploring the Frontier of AI: The "Tree of Thoughts" Framework
      • Toolformer: Revolutionising Language Models with API Integration - An Analysis
      • TaskMatrix.AI: Bridging Foundational AI Models with Specialised Systems for Enhanced Task Completion
      • Unleashing the Power of LLMs in API Integration: The Rise of Gorilla
      • Andrew Ng's presentation on AI agents
      • Making AI accessible with Andrej Karpathy and Stephanie Zhan
  • Regulation and Ethics
    • Regulation and Ethics
      • Privacy
      • Detecting AI Generated content
      • Navigating the IP Maze in AI: The Convergence of Blockchain, Web 3.0, and LLMs
      • Adverse Reactions to generative AI
      • Navigating the Ethical Minefield: The Challenge of Security in Large Language Models
      • Navigating the Uncharted Waters: The Risks of Autonomous AI in Military Decision-Making
  • DISRUPTION
    • Data Architecture
      • What is a data pipeline?
      • What is Reverse ETL?
      • Unstructured Data and Generatve AI
      • Resource Description Framework (RDF)
      • Integrating generative AI with the Semantic Web
    • Search
      • BM25 - Search Engine Ranking Function
      • BERT as a reranking engine
      • BERT and Google
      • Generative Engine Optimisation (GEO)
      • Billion-scale similarity search with GPUs
      • FOLLOWIR: Evaluating and Teaching Information Retrieval Models to Follow Instructions
      • Neural Collaborative Filtering
      • Federated Neural Collaborative Filtering
      • Latent Space versus Embedding Space
      • Improving Text Embeddings with Large Language Models
    • Recommendation Engines
      • On Interpretation and Measurement of Soft Attributes for Recommendation
      • A Survey on Large Language Models for Recommendation
      • Model driven recommendation systems
      • Recommender AI Agent: Integrating Large Language Models for Interactive Recommendations
      • Foundation Models for Recommender Systems
      • Exploring the Impact of Large Language Models on Recommender Systems: An Extensive Review
      • AI driven recommendations - harming autonomy?
    • Logging
      • A Taxonomy of Anomalies in Log Data
      • Deeplog
      • LogBERT: Log Anomaly Detection via BERT
      • Experience Report: Deep Learning-based System Log Analysis for Anomaly Detection
      • Log-based Anomaly Detection with Deep Learning: How Far Are We?
      • Deep Learning for Anomaly Detection in Log Data: A Survey
      • LogGPT
      • Adaptive Semantic Gate Networks (ASGNet) for log-based anomaly diagnosis
  • Infrastructure
    • The modern data centre
      • Enhancing Data Centre Efficiency: Strategies to Improve PUE
      • TCO of NVIDIA GPUs and falling barriers to entry
      • Maximising GPU Utilisation with Kubernetes and NVIDIA GPU Operator
      • Data Centres
      • Liquid Cooling
    • Servers and Chips
      • The NVIDIA H100 GPU
      • NVIDIA H100 NVL
      • Lambda Hyperplane 8-H100
      • NVIDIA DGX Servers
      • NVIDIA DGX-2
      • NVIDIA DGX H-100 System
      • NVLink Switch
      • Tensor Cores
      • NVIDIA Grace Hopper Superchip
      • NVIDIA Grace CPU Superchip
      • NVIDIA GB200 NVL72
      • Hopper versus Blackwell
      • HGX: High-Performance GPU Platforms
      • ARM Chips
      • ARM versus x86
      • RISC versus CISC
      • Introduction to RISC-V
    • Networking and Connectivity
      • Infiniband versus Ethernet
      • NVIDIA Quantum InfiniBand
      • PCIe (Peripheral Component Interconnect Express)
      • NVIDIA ConnectX InfiniBand adapters
      • NVMe (Non-Volatile Memory Express)
      • NVMe over Fabrics (NVMe-oF)
      • NVIDIA Spectrum-X
      • NVIDIA GPUDirect
      • Evaluating Modern GPU Interconnect
      • Scalable Hierarchical Aggregation and Reduction Protocol (SHARP)
      • Next-generation networking in AI environments
      • NVIDIA Collective Communications Library (NCCL)
    • Data and Memory
      • NVIDIA BlueField Data Processing Units (DPUs)
      • Remote Direct Memory Access (RDMA)
      • High Bandwidth Memory (HBM3)
      • Flash Memory
      • Model Requirements
      • Calculating GPU memory for serving LLMs
      • Transformer training costs
      • GPU Performance Optimisation
    • Libraries and Complements
      • NVIDIA Base Command
      • NVIDIA AI Enterprise
      • CUDA - NVIDIA GTC 2024 presentation
      • RAPIDs
      • RAFT
    • Vast Data Platform
      • Vast Datastore
      • Vast Database
      • Vast Data Engine
      • DASE (Disaggregated and Shared Everything)
      • Dremio and VAST Data
    • Storage
      • WEKA: A High-Performance Storage Solution for AI Workloads
      • Introduction to NVIDIA GPUDirect Storage (GDS)
        • GDS cuFile API
      • NVIDIA Magnum IO GPUDirect Storage (GDS)
      • Vectors in Memory
Powered by GitBook
LogoLogo

Continuum - Accelerated Artificial Intelligence

  • Continuum Website
  • Axolotl Platform

Copyright Continuum Labs - 2023

On this page
  • The L40S GPU - Introduction
  • Performance Specifications
  • Comparable Specifications
  • Applications and Use Cases
  • Performance Metrics
  • Conclusion
  • Specification Table

Was this helpful?

  1. INFERENCE
  2. Why is inference important?

NVIDIA L40S GPU

Low cost inference

PreviousTorchscriptNextTriton Inference Server - Introduction

Last updated 11 months ago

Was this helpful?

The L40S GPU - Introduction

The NVIDIA L40S GPU was launched late 2022. It is a powerful and versatile compute accelerator designed to meet the growing demands of AI, machine learning, and graphics-intensive applications in the data centre.

As the world becomes increasingly data-driven and AI-powered, there is a pressing need for high-performance computing solutions that can handle complex workloads efficiently and cost-effectively.

NVIDIA has developed the L40S to address this need, providing a solution that balances performance, flexibility, and scalability.

The GPU costs about $US12,000

The L40S is built on the NVIDIA Ada Lovelace architecture, which brings significant improvements in performance and efficiency compared to previous generations.

It is equipped with advanced features such as fourth-generation Tensor Cores, third-generation RT Cores, and the Transformer Engine, making it well-suited for a wide range of applications, including generative AI, large language model inference, 3D graphics, and video processing.

With its high-bandwidth memory, fast data transfer capabilities, and support for the latest display technologies, the L40S is a compelling choice for data centres looking to accelerate their AI and graphics workloads.

Performance Specifications

Tensor Performance

TFLOPS stands for "Tera Floating Point Operations Per Second." It's a measure of a computer's performance, specifically in tasks that require many mathematical calculations.

The L40S GPU can perform 1,466 trillion floating-point operations per second when working with tensors, which are multi-dimensional arrays used in machine learning and AI.

RT Core Performance

RT Cores are specialised processors within the GPU designed to handle ray tracing, a technique used to create realistic lighting and shadows in computer graphics. The L40S GPU can perform 212 trillion floating-point operations per second when using its RT Cores.

Single-Precision Performance

Single-precision refers to a specific format for representing decimal numbers in computer memory. The L40S GPU can perform 91.6 trillion floating-point operations per second when working with single-precision numbers.

Fourth-Generation Tensor Cores

Tensor Cores are specialised processors designed to accelerate AI and machine learning tasks.

The fourth-generation Tensor Cores in the L40S GPU have hardware support for features like structural sparsity and optimised TF32 format, which help improve performance and efficiency in AI and data science workloads.

Third-Generation RT Cores

The third-generation RT Cores in the L40S GPU have improved capabilities compared to previous generations. They can handle ray tracing more efficiently, resulting in faster and more realistic rendering for applications like product design and architecture.

CUDA Cores

CUDA (Compute Unified Device Architecture) Cores are the main processing units within the GPU. They are responsible for executing general-purpose computations and are particularly well-suited for tasks that can be parallelized, such as 3D modeling and computer simulations.

Transformer Engine

The Transformer Engine is a technology in the L40S GPU that optimises the performance and memory usage of transformer-based neural networks, which are commonly used in natural language processing and other AI applications.

It automatically adjusts the precision of calculations to deliver faster results and more efficient use of memory.

DLSS 3

DLSS (Deep Learning Super Sampling) is an AI-based technology that improves the performance and visual quality of rendered images.

DLSS 3, supported by the L40S GPU, uses deep learning and hardware innovations to generate new frames, resulting in smoother and faster rendering with improved latency.

Memory and Bandwidth

The NVIDIA L40S GPU comes with 48GB of GDDR6 memory, which is a high-bandwidth memory solution designed for graphics-intensive applications.

This large amount of memory allows the GPU to handle complex AI models, large datasets, and high-resolution graphics without running into memory limitations.

The L40S also supports ECC (Error Correction Code), a feature that detects and corrects errors in the memory automatically.

ECC is crucial in mission-critical applications where data integrity is paramount, such as scientific simulations, financial analysis, and healthcare.

By ensuring that the data stored in the GPU memory remains accurate and uncorrupted, ECC helps maintain the reliability and stability of the system.

The memory bandwidth of 864 GB/s refers to the rate at which data can be read from or written to the GPU memory.

A higher memory bandwidth allows the GPU to access and process data more quickly, which is essential when dealing with large datasets or complex models. The L40S's high memory bandwidth enables it to efficiently handle demanding workloads, reducing the time required for data transfer and computation.

Transformer Engine

The Transformer Engine is a key feature of the L40S GPU that significantly enhances its performance in AI workloads, particularly those involving transformer-based neural networks.

Transformers have become the dominant architecture in natural language processing (NLP) and have found applications in various other domains, such as computer vision and speech recognition.

The Transformer Engine optimises the execution of transformer models by scanning the layers of the neural network and dynamically adjusting the precision of the computations.

It can automatically switch between FP8 (8-bit floating-point) and FP16 (16-bit floating-point) precisions based on the requirements of each layer.

This dynamic precision adjustment allows the GPU to perform computations more efficiently, reducing memory usage and accelerating both training and inference workflows.

By leveraging lower-precision arithmetic (FP8) where possible and using higher precision (FP16) when needed, the Transformer Engine strikes a balance between accuracy and performance.

This optimisation is particularly beneficial for inference tasks, where the L40S can process transformer models with high throughput and low latency, enabling real-time applications such as language translation, text generation, and sentiment analysis.

PCIe Gen4 Interface and Display Outputs

PCIe (Peripheral Component Interconnect Express) is a high-speed serial interface that connects the GPU to the host system.

This high-bandwidth connection allows for fast communication between the GPU and the CPU, as well as other system components, enabling efficient data transfer and minimising bottlenecks.

In addition to the PCIe interface, the L40S also features four DisplayPort 1.4a outputs.

DisplayPort is a digital display interface standard that supports high resolutions and refresh rates.

With four DisplayPort outputs, the L40S can drive multiple high-resolution displays simultaneously, making it suitable for applications that require multi-monitor setups, such as visualisation, data analysis, and content creation.

Physical Size and Power Consumption

The dual-slot form factor of the L40S refers to its physical size and the number of expansion slots it occupies in a server chassis.

The dimensions of 4.4" (height) x 10.5" (length) make it compatible with standard server racks and chassis designs.

The maximum power consumption of 350W indicates the peak power draw of the GPU under full load, and the 16-pin power connector ensures that the GPU receives sufficient power to operate at its maximum performance level.

Data Centre Readiness

The L40S is built for 24/7 enterprise data centre operations and is designed, built, tested, and supported by NVIDIA to ensure maximum performance, durability, and uptime.

It meets the latest data centre standards and is NEBS (Network Equipment-Building System) Level 3 ready, ensuring reliable operation in demanding environments.

The GPU also features secure boot with root of trust technology, providing an additional layer of security for data centres.

It supports NVIDIA virtual GPU (vGPU) software, allowing multiple virtual machines to share the GPU resources and enabling efficient utilisation in virtualised environments.

Comparable Specifications

Applications and Use Cases

Generative AI in Creative Industries

The L40S can be deployed in creative industries, such as advertising agencies, game development studios, and movie production houses, to accelerate generative AI tasks.

For example, an advertising agency can use the L40S to generate multiple variations of ad designs, slogans, or product descriptions based on input prompts, allowing them to explore a wide range of creative options quickly.

Similarly, game developers can leverage the L40S to generate realistic textures, 3D models, or even entire game levels procedurally, saving time and effort in the design process.

Large Language Model Inference in Customer Service

The L40S can be used to deploy large language models for real-time inference in customer service applications.

For instance, a company can train a language model on their product documentation, FAQs, and customer interaction history, and then use the L40S to power a conversational AI chatbot.

The chatbot can understand and respond to customer queries in natural language, providing instant support and reducing the workload on human customer service representatives. The high-performance Tensor cores and Transformer Engine of the L40S ensure fast and accurate responses, enhancing the customer experience.

Fine-tuning LLMs for Domain-Specific Applications

Organisations can use the L40S to fine-tune pre-trained large language models for specific domains or use cases.

For example, a healthcare provider can fine-tune an LLM on medical literature, patient records, and clinical guidelines to create a specialised model that can assist doctors in diagnosis, treatment planning, and patient communication.

The computational power of the L40S allows for efficient fine-tuning, even with limited amounts of domain-specific data, enabling organizations to adapt LLMs to their unique requirements.

Collaborative 3D Design in NVIDIA Omniverse Enterprise

The L40S can be deployed in design and engineering firms to accelerate collaborative 3D design workflows using NVIDIA Omniverse Enterprise.

Omniverse is a platform that enables real-time collaboration and simulation of 3D models across different software tools and geographic locations.

With the L40S, designers and engineers can work together seamlessly, iterating on complex 3D models, running simulations, and visualising designs in high fidelity.

The GPU's RT cores and CUDA cores provide the necessary performance to render realistic lighting, shadows, and materials in real-time, facilitating faster decision-making and improved design outcomes.

High-Quality Video Streaming and Content Delivery

The L40S can be utilised by video streaming platforms and content delivery networks to encode, transcode, and deliver high-quality video content to users.

With support for advanced video codecs like AV1, the L40S can efficiently compress video streams while maintaining high visual quality, reducing bandwidth requirements and improving the user experience.

The GPU's encoding and decoding capabilities enable real-time video processing, allowing for live streaming, on-demand video delivery, and interactive video applications.

Content providers can leverage the L40S to scale their video infrastructure, handling multiple concurrent streams and delivering seamless video playback to a large user base.

Performance Metrics

In terms of performance metrics, the L40S demonstrates impressive results in various benchmarks.

For example, in the Stable Diffusion image generation benchmark, the L40S can generate 82 images per minute at 512x512 resolution and 17 images per minute at 1024x1024 resolution.

In the Large Language Model inference benchmark, the L40S achieves a 1st token latency of 77ms for the Llama 2-7B model, 143ms for the Llama 2-13B model, and 669ms for the Llama 2-70B model, demonstrating its ability to handle complex language models efficiently.

Conclusion

The NVIDIA L40S GPU is a powerful and versatile solution for data centres seeking to accelerate their AI, machine learning, and graphics workloads.

With its advanced architecture, high-performance components, and support for the latest technologies, the L40S is well-positioned to meet the growing demands of these applications. Its ability to efficiently handle diverse workloads, from generative AI and large language model inference to 3D graphics and video processing, makes it a valuable asset for organizations across various industries.

As the world continues to embrace AI and data-driven solutions, the L40S GPU will play a role in enabling businesses to unlock the full potential of these technologies and stay competitive in the evolving landscape.

Specification Table

Specification

Details

Explanation

Data Centre Ready

Yes, with NEBS Level 3 readiness and secure boot with root of trust

The L40S GPU is designed for continuous (24/7) operations in enterprise data centres. It adheres to high standards including NEBS Level 3 for reliability and includes security features like secure boot and root of trust to protect data integrity.

GPU Architecture

NVIDIA Ada Lovelace Architecture

Uses the latest NVIDIA GPU architecture, providing advanced processing power and efficiency for complex computations.

GPU Memory

48GB GDDR6 with ECC

Offers a large and fast memory capacity with error-correcting code (ECC) to detect and correct data corruption, ensuring data reliability in critical applications.

Memory Bandwidth

864GB/s

The rate at which data can be read from or stored into the GPU memory, indicating high data throughput capabilities.

Interconnect Interface

PCIe Gen4 x16: 64GB/s bidirectional

The bandwidth and type of interface used for connecting the GPU to the motherboard, allowing high-speed data transfer.

CUDA® Cores

18,176

The number of CUDA cores indicates the parallel processing power of the GPU, essential for handling multiple operations simultaneously.

RT Cores / Tensor Cores

142 Third-Generation RT Cores / 568 Fourth-Generation Tensor Cores

RT cores are specialized for ray tracing operations, enhancing realistic lighting and shadows in 3D environments. Tensor cores are designed for deep learning operations, speeding up AI computations.

Core Performance

Various TFLOPS and TOPS metrics

Measures the floating point (TFLOPS) and tensor operations per second (TOPS) performance, key indicators of the GPU’s capability in handling different types of computational loads from graphics rendering to AI processing.

Form Factor

4.4" (H) x 10.5" (L), dual slot

Physical dimensions and slot requirement for the GPU, indicating the amount of space needed within the computer chassis.

Display Ports

4x DisplayPort 1.4a

Type and number of available ports for connecting displays, supporting high-resolution and multiple monitor setups.

Max Power Consumption

350W

The maximum amount of power the GPU consumes, important for understanding the energy requirements and thermal design considerations.

Power Connector

16-pin

The type of power connector used, indicating compatibility with power supplies and the required power delivery for stable operation.

Thermal

Passive

The cooling method used by the GPU, with passive cooling typically involving no fans, relying instead on heat sinks to dissipate heat, useful in environments where noise reduction is critical.

Virtual GPU (vGPU) Software Support

Yes

Indicates support for virtualization, allowing the GPU’s resources to be divided for use by multiple virtual machines, enhancing flexibility in multi-user environments.

vGPU Profiles Supported

Refer to the virtual GPU licensing guide

Details which specific virtual GPU configurations are available and supported, important for setup and management in virtualized environments.

NVENC / NVDEC

3x l 3x (includes AV1 encode and decode)

Specifies the capabilities of NVIDIA's hardware-accelerated video encoding (NVENC) and decoding (NVDEC) including support for the latest AV1 codec, enhancing video processing tasks.

Secure Boot With Root of Trust

Yes

Confirms the presence of mechanisms that verify the integrity of the hardware and software, preventing unauthorized changes to the system.

NEBS Ready Level 3

Yes

Indicates compliance with stringent standards for reliability and robustness required for operation in demanding environments, such as telecommunications facilities.

MIG Support

No

Machine Instance GPU (MIG) support status, which if supported, would allow the physical GPU to be partitioned into smaller instances, each capable of running isolated workloads.

NVIDIA® NVLink® Support

No

Indicates whether the GPU can be connected to other GPUs using NVIDIA’s NVLink technology, which would enable high-speed data sharing enhancing performance in multi-GPU configurations.

The L40S supports , which is the latest generation of the PCIe standard, providing a data transfer rate of up to 64 GB/s in each direction (bidirectional).

PCIe Gen4
The NVIDIA L40S GPU
Comparing the specifications of the L40S and the 80GB version of the A100 and H100
Page cover image