NVIDIA L40S
Bare Metal Servers
Command the ultimate universal data center GPU starting at $1656/mo. Featuring 48GB of GDDR6 ECC memory per card, the Ada Lovelace-powered L40S shatters benchmarks across AI inference, 3D rendering, and Omniverse simulations.
Configure Your Universal AI Node
No Servers Match Your Exact Filters
We have thousands of servers in stock! Try broadening your location, adjusting your CPU requirements, or expanding your budget.
The Universal Data Center GPU.
The NVIDIA L40S is the most powerful universal GPU for the data center, delivering end-to-end acceleration for next-generation AI inference, graphics, and video workloads.
48GB
GDDR6 with ECC
Deliver uncompromising reliability for enterprise deployments. Error Correction Code (ECC) prevents corruption during massive LLM training and heavy digital twin simulations.
864 GB/s
Memory Bandwidth
Feed data hungry AI models instantly. High-bandwidth memory operating over a 384-bit interface ensures your 18,176 CUDA cores are utilized to their absolute maximum.
4th Gen
Tensor Cores
Features FP8 support and the powerful Transformer Engine, accelerating complex generative AI algorithms and delivering massive leap in inference performance.
3rd Gen
RT Cores
Dramatically accelerate rendering of photorealistic environments, 3D designs, and real-time ray-traced metaverse applications in NVIDIA Omniverse.
Breakthrough Performance for
AI and Visual Computing
The NVIDIA L40S bridges the gap between deep learning and stunning visual computing. Unlike GPUs built strictly for AI or strictly for graphics, the L40S excels at both, making it the most cost-effective hardware for mixed enterprise workloads.
-
Transformer Engine Optimization
Natively supports FP8 data formats. It dynamically optimizes the precision of AI models, slashing inference latency and boosting throughput for large language models (LLMs).
-
Advanced Video & Image Rendering
Equipped with three 8th-generation NVENC hardware encoders featuring native AV1 encoding, allowing media platforms to stream higher quality video at significantly lower bitrates.
-
PCIe Gen 4/5 Efficiency
The L40S operates highly efficiently over standard PCIe lanes. It delivers H100-tier performance in specific generative AI tasks without requiring the massive networking overhead of proprietary NVLink bridges.
In-Stock & Ready for Global Deployment
Built for the Omniverse & AI
Harness the unrivaled flexibility of the L40S across demanding enterprise, media, and simulation environments.
Generative AI & LLMs
Train and fine-tune complex models efficiently. The L40S is a powerhouse for NLP text generation, AI image synthesis, and running responsive, high-throughput inference APIs for end users.
Omniverse & Digital Twins
Pioneered for the industrial metaverse. Process hyper-realistic, physically accurate 3D simulations and factory digital twins in real-time utilizing massive 3rd-Gen RT cores and 48GB VRAM buffers.
Computer Vision & Video
Ingest, decode, and analyze hundreds of live video streams concurrently. Ideal for smart city infrastructure, automated quality control, and massive-scale AV1 media transcoding farms.
Scale from 1x to 8x Multi-GPU Clusters
Scale your universal compute horizontally. iDatam allows you to deploy highly dynamic 1x, 2x, and massive 8x NVIDIA L40S chassis seamlessly paired with premium AMD EPYC and multi-core Intel Xeon Gold architectures.
384GB
Max Cluster VRAM
4x 25G
High-Speed Links
NVMe
Enterprise Storage
$1656
Starting Price / Mo
Deploy Globally
Provision your universal L40S compute nodes instantly in secure data center architectures, strategically mapped across high-availability corridors.
- Italy (Arezzo / Bergamo) Ready
- Sweden (Stockholm / Falkenberg) Ready
- United Kingdom (London) Ready
- Netherlands (Amsterdam / Hague) Ready
NVIDIA L40S FAQs
Common questions about deploying the Ada Lovelace universal data center GPU.