NVIDIA L40S 48GB GDDR6 PCIe GPU Accelerator

The NVIDIA L40S 48GB GDDR6 PCIe GPU Accelerator is built on the NVIDIA Ada Lovelace architecture, delivering a substantial leap in compute density and energy efficiency for modern enterprise and data center environments. With 142 second-generation RT Cores, 568 fourth-generation Tensor Cores, and 18,176 CUDA cores, the L40S is engineered to handle the most demanding AI inference, training, and graphics workloads simultaneously within a single-slot, full-height, full-length form factor. The GPU connects via PCIe Gen 4 x16 and supports NVLink bridge configurations for multi-GPU scaling, making it highly adaptable to a wide range of server chassis and rack configurations.

At the heart of the L40S is 48GB of GDDR6 memory operating on a 384-bit bus with a memory bandwidth of 864 GB/s, providing the headroom required for large language models, high-resolution rendering pipelines, and complex simulation workloads. The card delivers up to 91.6 TFLOPS of FP32 performance and up to 362 TOPS of INT8 inference throughput with sparsity, while the Transformer Engine with FP8 precision enables accelerated training and inference of modern generative AI models. The L40S operates within a 350W TDP envelope, with power supplied through dual 8-pin PCIe connectors, and supports passive cooling for compatibility with airflow-managed data center environments.

Designed explicitly for enterprise data center deployment, the L40S differentiates itself from consumer-grade accelerators through its support for ECC memory, professional-grade reliability features, and NVIDIA's suite of enterprise software stacks including CUDA, TensorRT, and Omniverse. The Gigabyte-branded L40S (MPN: 900-2G133-0040-000) targets organizations in sectors including financial services, healthcare, media and entertainment, telecommunications, and cloud service provision, where sustained, mixed-workload GPU utilization at scale is a core operational requirement. Omnixon Global makes this accelerator available to enterprise buyers across the UAE, GCC, EMEA, and APAC regions.

Ideal for

• Large-scale generative AI model training and fine-tuning, including large language models and diffusion-based image generation, leveraging FP8 Tensor Core throughput
• High-throughput AI inference serving for real-time applications such as recommendation engines, natural language processing APIs, and computer vision pipelines
• 3D visualization, ray tracing, and GPU-accelerated rendering for media production, architecture, and digital twin development using RT Core and CUDA resources
• Scientific simulation and high-performance computing workloads in industries such as energy, life sciences, and computational fluid dynamics
• Virtual workstation and cloud graphics delivery for demanding professional applications via GPU virtualization with NVIDIA vGPU software
• Telecommunications and edge AI acceleration in centralized data center nodes supporting 5G network analytics, signal processing, and real-time inference

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.