NVIDIA L40S 48GB PCIe Gen4 GPU
Brand: NVIDIA | Category: GPUs
SKU: NVID-9002G1330050000 | Part #: 900-2G133-0050-000 | MPN: 900-2G133-0050-000
Contact for Pricing — Request a Quote
About the NVIDIA L40S 48GB PCIe Gen4 GPU
The NVIDIA L40S is a full-height, full-length PCIe Gen4 accelerator built on the Ada Lovelace GPU architecture, featuring 142 third-generation Tensor Cores and 142 RT Cores across 18,176 CUDA cores. With 48GB of GDDR6 ECC memory and a 864 GB/s memory bandwidth, the L40S is engineered to handle the simultaneous demands of generative AI inference, large language model serving, 3D visualization, and video processing within a single card—eliminating the need to dedicate separate GPU pools to distinct workload types. The 300W TDP and passive cooling design integrate cleanly into standard server platforms from Dell, HPE, Lenovo, Supermicro, and others without requiring custom thermal solutions, making it broadly deployable across existing data center infrastructure.
Ideal for
- Large language model inference serving for enterprise chatbots and copilot applications requiring low-latency token generation
- Generative AI image and video synthesis pipelines including diffusion model workloads such as Stable Diffusion XL
- GPU-accelerated virtual workstation delivery (NVIDIA RTX Virtual Workstation) for remote CAD, BIM, and media production teams
- Real-time ray-traced 3D visualization and digital twin rendering in industrial simulation environments
- Multi-instance GPU (MIG) partitioned inference serving to efficiently host multiple smaller AI models concurrently for SaaS platforms
- Video transcoding and AI-enhanced upscaling pipelines for broadcast, streaming, and media asset management at scale
Technical specifications
| Manufacturer | NVIDIA |
| GPU Architecture | Ada Lovelace (AD102) |
| CUDA Cores | 18,176 |
| Third-Gen Tensor Cores | 568 |
| RT Cores (3rd Gen) | 142 |
| FP32 Performance | 91.6 TFLOPS |
| TF32 Tensor Core Performance | 183 TFLOPS (sparsity: 366 TFLOPS) |
| FP16 Tensor Core Performance | 362.05 TFLOPS (sparsity: 724.1 TFLOPS) |
| INT8 Tensor Core Performance | 724.1 TOPS (sparsity: 1,457.9 TOPS) |
| Memory Capacity | 48 GB GDDR6 ECC |
| Memory Bandwidth | 864 GB/s |
| Memory Interface | 384-bit |
| PCIe Interface | PCIe Gen4 x16 |
| NVLink / NVLink Bridge | NVLink 4.0 (NVLink Bridge supported for 2-GPU configurations) |
| TDP (Thermal Design Power) | 300 W |
| Form Factor | Full-Height Full-Length (FHFL), Dual-Slot |
| Cooling | Passive (requires server airflow) |
| Display Outputs | 4× DisplayPort 1.4a |
| Multi-Instance GPU (MIG) | Not supported (MIG reserved for H/A-series); supports Multi-GPU via NVLink |
| NVIDIA vGPU Software Support | RTX Virtual Workstation (vWS), Virtual PC (vPC), Virtual Compute Server (vCS) |
| ECC Memory | Yes, GDDR6 ECC |
| NVENC / NVDEC Engines | 3× NVENC, 3× NVDEC, 1× AV1 Encode |
| Maximum GPU Board Power Connectors | 16-pin PCIe 5.0 (or dual 8-pin adapter) |
| Operating Temperature | 0°C to 45°C (ambient) |
| Launch Generation | 2024 Q3 broad platform availability |
Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.
Technical Specifications
| Brand | NVIDIA |
| Category | GPUs |
| SKU | NVID-9002G1330050000 |
| Part Number | 900-2G133-0050-000 |
| Condition | New |
| GPU Architecture | Ada Lovelace (AD102) |
| CUDA Cores | 18,176 |
| Third-Gen Tensor Cores | 568 |
| RT Cores (3rd Gen) | 142 |
| FP32 Performance | 91.6 TFLOPS |
| TF32 Tensor Core Performance | 183 TFLOPS (sparsity: 366 TFLOPS) |
| FP16 Tensor Core Performance | 362.05 TFLOPS (sparsity: 724.1 TFLOPS) |
| INT8 Tensor Core Performance | 724.1 TOPS (sparsity: 1,457.9 TOPS) |
| Memory Capacity | 48 GB GDDR6 ECC |
| Memory Bandwidth | 864 GB/s |
| Memory Interface | 384-bit |
| PCIe Interface | PCIe Gen4 x16 |
| NVLink / NVLink Bridge | NVLink 4.0 (NVLink Bridge supported for 2-GPU configurations) |
| TDP (Thermal Design Power) | 300 W |
| Form Factor | Full-Height Full-Length (FHFL), Dual-Slot |
| Cooling | Passive (requires server airflow) |
| Display Outputs | 4× DisplayPort 1.4a |
| Multi-Instance GPU (MIG) | Not supported (MIG reserved for H/A-series); supports Multi-GPU via NVLink |
| NVIDIA vGPU Software Support | RTX Virtual Workstation (vWS), Virtual PC (vPC), Virtual Compute Server (vCS) |
| ECC Memory | Yes, GDDR6 ECC |
| NVENC / NVDEC Engines | 3× NVENC, 3× NVDEC, 1× AV1 Encode |
| Maximum GPU Board Power Connectors | 16-pin PCIe 5.0 (or dual 8-pin adapter) |
| Operating Temperature | 0°C to 45°C (ambient) |
| Launch Generation | 2024 Q3 broad OEM platform availability |
Frequently Asked Questions about NVIDIA L40S 48GB PCIe Gen4 GPU
What does the NVIDIA L40S 48GB PCIe Gen4 GPU do?
The NVIDIA L40S 48GB PCIe Gen4 GPU accelerates AI/ML training, inference, scientific HPC, and virtualization (vGPU) workloads. Typical deployments include LLM training clusters, computer-vision pipelines, financial risk modeling, and rendering farms.
What are the headline specs of the NVIDIA L40S 48GB PCIe Gen4 GPU?
Key specifications for the NVIDIA L40S 48GB PCIe Gen4 GPU: new condition; manufacturer NVIDIA; gpu architecture Ada Lovelace (AD102); cuda cores 18,176; third-gen tensor cores 568; rt cores (3rd gen) 142; fp32 performance 91.6 TFLOPS. Manufacturer part number 900-2G133-0050-000. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.
Is the NVIDIA L40S 48GB PCIe Gen4 GPU compatible with my infrastructure?
The NVIDIA L40S 48GB PCIe Gen4 GPU requires a PCIe Gen4 or Gen5 x16 slot, server power adequate for the card's TDP, and CUDA/ROCm driver support in your hypervisor or bare-metal OS. Sales engineering will confirm chassis fit (1U/2U/4U), PCIe lane count, and PSU headroom before quoting.