L40S GPU

The NVIDIA L40S is a high-performance data center GPU built on the Ada Lovelace architecture, positioned as a workstation-class accelerator for enterprise AI and visual computing workloads. Released in 2023, it sits below NVIDIA's server-focused H100 and newer GB300 series in the data center hierarchy, targeting organizations that need substantial AI inference capabilities combined with graphics rendering performance in a single card. The L40S features 48GB of GDDR6 memory with ECC protection, 18,176 CUDA cores, and 568 fourth-generation Tensor Cores that deliver 362 TFLOPS of FP16 performance and 1,466 TFLOPS of FP8 tensor performance. Its 864 GB/s memory bandwidth supports memory-intensive workloads, while the inclusion of third-generation RT Cores provides 212 TFLOPS of ray tracing performance. The GPU incorporates the Transformer Engine for optimized large language model processing and supports DLSS 3 for accelerated graphics workloads. In cloud deployments, the L40S serves dual-purpose scenarios where both AI inference and graphics rendering capabilities are needed. Its design for 24/7 data center operations with secure boot and root of trust technology makes it suitable for enterprise environments running mixed workloads including generative AI applications, LLM inference, and professional visualization tasks through platforms like NVIDIA Omniverse.