pchardware.org
Categories

NVIDIA · 2022

NVIDIA L40 GPU accelerator

The NVIDIA L40 is a professional GPU accelerator based on the Ada Lovelace architecture, designed for data center workloads.

The NVIDIA L40 GPU accelerator, released in October 2022, is built on the Ada Lovelace architecture using a 4 nm process. It features 18176 shading units, 142 ray tracing cores, and 568 Tensor/AI cores. The GPU operates at a base clock of 1155 MHz and a boost clock of 2475 MHz. It is equipped with 48 GB of GDDR6 memory on a 384-bit bus, providing a memory bandwidth of 864 GB/s. The board has a TDP of 300 W and uses a single 16-pin 12VHPWR power connector. It connects via PCIe 4.0 x16 and has no display outputs, indicating it is intended for compute and AI acceleration rather than graphics output. It supports DirectX 12 Ultimate, Vulkan 1.3, and OpenGL 4.6. The launch MSRP was $7999 USD.

Specifications

Architecture unverified Ada Lovelace
GPU chip unverified AD102
Process node unverified 4 nm
Shading units unverified 18176
Ray tracing cores unverified 142
Tensor / AI cores unverified 568
Base clock unverified 1155 MHz
Boost clock unverified 2475 MHz
Memory size unverified 48 GB
Memory type unverified GDDR6
Memory bus unverified 384 bit
Memory bandwidth unverified 864 GB/s
Board power (TDP) unverified 300 W
Power connectors unverified 1x 16-pin (12VHPWR)
Bus interface unverified PCIe 4.0 x16
Display outputs unverified No display outputs
DirectX support unverified 12 Ultimate
Vulkan support unverified 1.3
OpenGL support unverified 4.6
Release date unverified 2022-10-13
Release year unverified 2022
Launch MSRP unverified 7999 USD