Nvidia A2 GPU

Performance

The A2 is an entry-level data center GPU powered by the 8nm GA107 processor, the same processor on the GeForce RTX 3050 GPU. Even though this is a half-length, half-height, single-slot form factor GPU, it packs a lot of power. Featuring 1280 shading units, 40 texture mapping units, and 32 ROPs, the card also supports 40 tensor cores to increase the speed of machine learning applications, and 10 raytracing acceleration cores (RTcores). Edge and entry-level servers with NVIDIA A2 Tensor Core GPUs provide up to 20X more inference performance than CPU-only servers, immediately upgrading any server to handle modern AI.

Memory

Fueled by Ampere architecture, NVIDIA has combined 16 GB of GDDR6 memory with a 128-bit memory interface clocking in at 1563 MHz, or 12.5 Gbps. The GPU runs at a frequency of 1440 MHz, which may be increased to up to 1770 MHz.

Features

Since the NVIDIA A2 GPU is a single-slot card designed for server applications, it does not have any display connectivity ports nor does it require any additional power connectors. Instead, the GPU is mounted on the server using a PCI-Express 4.0 x8 interface and is powered directly through the PCIs slot. As a passively cooled PCIe card, it relies on a bidirectional heat sink that can draw airflow from either the left or right. The device is designed for entry-level servers with space and thermal constraints, offloading a TDP ranging from 40W to 60W.

Summary

Built on an 8nm process, the NVIDIA A2 GPU features 1280 CUDA cores, 40 Tensor cores, and 16GB of GDDR6 memory. With the latest Ampere architecture and PCIe Gen 4 capabilities, this card meets the demands of high-performance workloads to deliver leading inference performance across edge, data center, and cloud with high efficiency.