900-2G133-00A0-000

• NVIDIA L20, L-Series, Ada Lovelace Architecture on the AD102 GPU, 5 Nm Lithography
• 48 GB GDDR6 memory, 384-Bit bus, 864 GB/S bandwidth, 18 Gbps Memory Clock
• 11, 776 CUDA cores, 368 Tensor cores, 92 RT Cores
• Core Clock 1305 MHz Boosting to 2520 MHz
• Peak Performance 59.4 TFLOPS FP32 and FP16, 9.2 TFLOPS FP64
• PCIe 4.0 x16 System interface, Dual-Slot Form factor, 10.5 in X 4.4 in
• Passive cooling, 275 W TDP, One 16-Pin Power connector, 600 W Recommended PSU
• API Support: DirectX 12 Ultimate, OpenGL 4.6, OpenCL 3, Vulkan 1

The NVIDIA 900-2G133-00A0-000 is an L20 professional graphics card built on the Ada Lovelace architecture with the AD102 GPU on a 5 nm process. It carries 48 GB of GDDR6 memory on a 384-bit bus delivering 864 GB/s of bandwidth at an 18 Gbps memory clock. The GPU provides 11, 776 CUDA cores, 368 Tensor cores, and 92 RT cores, boosting to 2520 MHz from a 1305 MHz core clock, and reaches 59.4 TFLOPS at FP32 and FP16 and 9.2 TFLOPS at FP64. The card is a passively cooled dual-slot PCIe 4.0 x16 board rated at 275 W TDP with a single 16-pin power connector.

A passively cooled 275 W dual-slot accelerator with 48 GB of memory is built for rack servers that supply their own front-to-back airflow, not for workstations with case fans. The 48 GB frame buffer and Tensor cores target AI inference on mid-to-large models, generative AI serving, and video transcoding and rendering workloads, while the RT cores also support virtual workstation and visualisation duty. Typical users include cloud and hosting providers, enterprise AI inference platforms, VDI and virtual workstation deployments, and media and rendering farms. Confirm chassis airflow and 16-pin power availability before installation, as the passive heatsink depends on server-supplied cooling.