NVIDIA L40S
NVIDIA Graphics Card
The NVIDIA L40S is a universal data centre GPU built on the Ada Lovelace architecture with 18,176 CUDA cores and 48GB of GDDR6 ECC memory delivering 864 GB/s of bandwidth. It combines strong AI throughput (up to 1,466 TFLOPS of FP8 Tensor performance) with a full graphics and video encode/decode pipeline, making it one of the few single cards able to handle AI inference, model fine-tuning, 3D rendering and video transcoding equally well. The PCIe 4.0 dual-slot passively cooled design supports high-density deployment in standard rack servers.
Specifications
| Architecture | NVIDIA Ada Lovelace architecture (AD102) 4th-generation Tensor Cores 3rd-generation RT Cores |
|---|---|
| Compute Cores | 18,176 NVIDIA CUDA Cores 568 Tensor Cores, 142 RT Cores |
| AI Performance | Up to 1,466 TFLOPS FP8 Tensor performance |
| GPU Memory | 48GB GDDR6 with ECC 864 GB/s memory bandwidth |
| System Interface | PCIe 4.0 x16 |
| Media Engines | 3x NVENC encoders (AV1 support) 3x NVDEC decoders |
| Virtualisation | NVIDIA vGPU software support |
| Power | 350W max power consumption |
| Cooling & Form Factor | Passive cooling (chassis airflow) Dual-slot full-height full-length, for rack servers |
| Use Cases | AI inference and fine-tuning, generative AI, virtual desktop infrastructure, cloud rendering, video transcoding |