Nvidia A2 is the green company’s surprise graphics accelerator for servers.
Nvidia A2: Low-profile server graphics accelerator announced
Nvidia is targeting the server segment with its new A2 graphics accelerators, which offer many advantages for customers in this segment with their relatively small size and reduced power consumption.
With low-end features in the graphics accelerator category, the Nvidia A2 would replace the well-known Nvidia T4 series of accelerators used for inference tasks with the same single-slot HHHL card format.
It appears that the launch of this card will be the last from the green company for the server segment, where a low-end variant with the Ampere architecture was missing.
The Nvidia A2 would use a low-end variant of the GA107 GPU. This GPU would have 1,280 CUDA cores (and 40 Tensor cores). This is half the power envelope of the GA107 GPU, which is clearly meant to optimize power consumption. The A2 consumes only 60 W but can be configured to consume only 42 W, which sounds very tempting for the target segment.
VRAM memory capacity is 16 GB with a 128-bit memory bus that operates at a speed of 12.5 Gbps. This corresponds to a memory bandwidth of 200 GB/second.
According to the green company, the A2 accelerator graphics card is available now. No price was given, but it should be available from the usual OEM partners.
At GTC today, Jensen Huang delivered a keynote address describing his vision for a wide range of areas including accelerated computing, data center architecture, AI, robotics, Omniverse avatars, and digital twins. During his presentation, Jensen Huang introduced a number of recent announcements and demos, including NVIDIA Omniverse Avatar and NVIDIA Omniverse Replicator, and showcased NVIDIA Omniverse, a virtual world simulation and collaboration platform for 3D workflows. Omniverse and how Omniverse combines NVIDIA’s technologies.