NVIDIA A800 is a graphics processor designed to work in the most demanding data center applications. This model was developed in accordance with the new US export control rules as an alternative to the NVIDIA A100 GPU for customers in China.
Key Features
The A800 has 6912 shading units, 432 texturing units and 160 ROPs. The processor has 432 tensor cores, which increase the speed of machine learning applications.
The GPU runs at 765 MHz, which can be upgraded to 1410 MHz. The A800 is equipped with a massive 40 GB HBM2e memory connected via a 5120-bit memory interface.
Power and performance
The A800 is powered by an 8-pin EPS power connector, and the maximum power consumption is 250 watts. Despite this, the A800 is designed as a dual-slot processor, which makes it compatible with various systems.
Physical dimensions
The length of the A800 is 267 mm in length and the width is 111 mm. The GPU is equipped with a dual-slot cooling system to control heat generation during intensive operation.
Connectivity options
The A800 is not designed to connect monitors, instead the GPU connects to the rest of the system using the PCI-Express 4.0 x16 interface.
Efficiency
The NVIDIA A800 runs at the same speed as the NVIDIA A100. The data transfer rate may differ only when using NVLink — combining several cards into one.
The A800 provides the following performance: 6.8 FP64/13.7 FP64 Tensor TFLOPS, as well as 437 BF16/FP16 (with sparsity).
Applications of NVIDIA A800
A800 GPUs, like the A100 and other modern NVIDIA models, are suitable for deep learning, interface optimization, high-performance computing and data analysis.